xref: /openbmc/linux/arch/x86/kvm/emulate.c (revision 7cc39531)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /******************************************************************************
3  * emulate.c
4  *
5  * Generic x86 (32-bit and 64-bit) instruction decoder and emulator.
6  *
7  * Copyright (c) 2005 Keir Fraser
8  *
9  * Linux coding style, mod r/m decoder, segment base fixes, real-mode
10  * privileged instructions:
11  *
12  * Copyright (C) 2006 Qumranet
13  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
14  *
15  *   Avi Kivity <avi@qumranet.com>
16  *   Yaniv Kamay <yaniv@qumranet.com>
17  *
18  * From: xen-unstable 10676:af9809f51f81a3c43f276f00c81a52ef558afda4
19  */
20 
21 #include <linux/kvm_host.h>
22 #include "kvm_cache_regs.h"
23 #include "kvm_emulate.h"
24 #include <linux/stringify.h>
25 #include <asm/debugreg.h>
26 #include <asm/nospec-branch.h>
27 #include <asm/ibt.h>
28 
29 #include "x86.h"
30 #include "tss.h"
31 #include "mmu.h"
32 #include "pmu.h"
33 
34 /*
35  * Operand types
36  */
37 #define OpNone             0ull
38 #define OpImplicit         1ull  /* No generic decode */
39 #define OpReg              2ull  /* Register */
40 #define OpMem              3ull  /* Memory */
41 #define OpAcc              4ull  /* Accumulator: AL/AX/EAX/RAX */
42 #define OpDI               5ull  /* ES:DI/EDI/RDI */
43 #define OpMem64            6ull  /* Memory, 64-bit */
44 #define OpImmUByte         7ull  /* Zero-extended 8-bit immediate */
45 #define OpDX               8ull  /* DX register */
46 #define OpCL               9ull  /* CL register (for shifts) */
47 #define OpImmByte         10ull  /* 8-bit sign extended immediate */
48 #define OpOne             11ull  /* Implied 1 */
49 #define OpImm             12ull  /* Sign extended up to 32-bit immediate */
50 #define OpMem16           13ull  /* Memory operand (16-bit). */
51 #define OpMem32           14ull  /* Memory operand (32-bit). */
52 #define OpImmU            15ull  /* Immediate operand, zero extended */
53 #define OpSI              16ull  /* SI/ESI/RSI */
54 #define OpImmFAddr        17ull  /* Immediate far address */
55 #define OpMemFAddr        18ull  /* Far address in memory */
56 #define OpImmU16          19ull  /* Immediate operand, 16 bits, zero extended */
57 #define OpES              20ull  /* ES */
58 #define OpCS              21ull  /* CS */
59 #define OpSS              22ull  /* SS */
60 #define OpDS              23ull  /* DS */
61 #define OpFS              24ull  /* FS */
62 #define OpGS              25ull  /* GS */
63 #define OpMem8            26ull  /* 8-bit zero extended memory operand */
64 #define OpImm64           27ull  /* Sign extended 16/32/64-bit immediate */
65 #define OpXLat            28ull  /* memory at BX/EBX/RBX + zero-extended AL */
66 #define OpAccLo           29ull  /* Low part of extended acc (AX/AX/EAX/RAX) */
67 #define OpAccHi           30ull  /* High part of extended acc (-/DX/EDX/RDX) */
68 
69 #define OpBits             5  /* Width of operand field */
70 #define OpMask             ((1ull << OpBits) - 1)
71 
72 /*
73  * Opcode effective-address decode tables.
74  * Note that we only emulate instructions that have at least one memory
75  * operand (excluding implicit stack references). We assume that stack
76  * references and instruction fetches will never occur in special memory
77  * areas that require emulation. So, for example, 'mov <imm>,<reg>' need
78  * not be handled.
79  */
80 
81 /* Operand sizes: 8-bit operands or specified/overridden size. */
82 #define ByteOp      (1<<0)	/* 8-bit operands. */
83 /* Destination operand type. */
84 #define DstShift    1
85 #define ImplicitOps (OpImplicit << DstShift)
86 #define DstReg      (OpReg << DstShift)
87 #define DstMem      (OpMem << DstShift)
88 #define DstAcc      (OpAcc << DstShift)
89 #define DstDI       (OpDI << DstShift)
90 #define DstMem64    (OpMem64 << DstShift)
91 #define DstMem16    (OpMem16 << DstShift)
92 #define DstImmUByte (OpImmUByte << DstShift)
93 #define DstDX       (OpDX << DstShift)
94 #define DstAccLo    (OpAccLo << DstShift)
95 #define DstMask     (OpMask << DstShift)
96 /* Source operand type. */
97 #define SrcShift    6
98 #define SrcNone     (OpNone << SrcShift)
99 #define SrcReg      (OpReg << SrcShift)
100 #define SrcMem      (OpMem << SrcShift)
101 #define SrcMem16    (OpMem16 << SrcShift)
102 #define SrcMem32    (OpMem32 << SrcShift)
103 #define SrcImm      (OpImm << SrcShift)
104 #define SrcImmByte  (OpImmByte << SrcShift)
105 #define SrcOne      (OpOne << SrcShift)
106 #define SrcImmUByte (OpImmUByte << SrcShift)
107 #define SrcImmU     (OpImmU << SrcShift)
108 #define SrcSI       (OpSI << SrcShift)
109 #define SrcXLat     (OpXLat << SrcShift)
110 #define SrcImmFAddr (OpImmFAddr << SrcShift)
111 #define SrcMemFAddr (OpMemFAddr << SrcShift)
112 #define SrcAcc      (OpAcc << SrcShift)
113 #define SrcImmU16   (OpImmU16 << SrcShift)
114 #define SrcImm64    (OpImm64 << SrcShift)
115 #define SrcDX       (OpDX << SrcShift)
116 #define SrcMem8     (OpMem8 << SrcShift)
117 #define SrcAccHi    (OpAccHi << SrcShift)
118 #define SrcMask     (OpMask << SrcShift)
119 #define BitOp       (1<<11)
120 #define MemAbs      (1<<12)      /* Memory operand is absolute displacement */
121 #define String      (1<<13)     /* String instruction (rep capable) */
122 #define Stack       (1<<14)     /* Stack instruction (push/pop) */
123 #define GroupMask   (7<<15)     /* Opcode uses one of the group mechanisms */
124 #define Group       (1<<15)     /* Bits 3:5 of modrm byte extend opcode */
125 #define GroupDual   (2<<15)     /* Alternate decoding of mod == 3 */
126 #define Prefix      (3<<15)     /* Instruction varies with 66/f2/f3 prefix */
127 #define RMExt       (4<<15)     /* Opcode extension in ModRM r/m if mod == 3 */
128 #define Escape      (5<<15)     /* Escape to coprocessor instruction */
129 #define InstrDual   (6<<15)     /* Alternate instruction decoding of mod == 3 */
130 #define ModeDual    (7<<15)     /* Different instruction for 32/64 bit */
131 #define Sse         (1<<18)     /* SSE Vector instruction */
132 /* Generic ModRM decode. */
133 #define ModRM       (1<<19)
134 /* Destination is only written; never read. */
135 #define Mov         (1<<20)
136 /* Misc flags */
137 #define Prot        (1<<21) /* instruction generates #UD if not in prot-mode */
138 #define EmulateOnUD (1<<22) /* Emulate if unsupported by the host */
139 #define NoAccess    (1<<23) /* Don't access memory (lea/invlpg/verr etc) */
140 #define Op3264      (1<<24) /* Operand is 64b in long mode, 32b otherwise */
141 #define Undefined   (1<<25) /* No Such Instruction */
142 #define Lock        (1<<26) /* lock prefix is allowed for the instruction */
143 #define Priv        (1<<27) /* instruction generates #GP if current CPL != 0 */
144 #define No64	    (1<<28)
145 #define PageTable   (1 << 29)   /* instruction used to write page table */
146 #define NotImpl     (1 << 30)   /* instruction is not implemented */
147 /* Source 2 operand type */
148 #define Src2Shift   (31)
149 #define Src2None    (OpNone << Src2Shift)
150 #define Src2Mem     (OpMem << Src2Shift)
151 #define Src2CL      (OpCL << Src2Shift)
152 #define Src2ImmByte (OpImmByte << Src2Shift)
153 #define Src2One     (OpOne << Src2Shift)
154 #define Src2Imm     (OpImm << Src2Shift)
155 #define Src2ES      (OpES << Src2Shift)
156 #define Src2CS      (OpCS << Src2Shift)
157 #define Src2SS      (OpSS << Src2Shift)
158 #define Src2DS      (OpDS << Src2Shift)
159 #define Src2FS      (OpFS << Src2Shift)
160 #define Src2GS      (OpGS << Src2Shift)
161 #define Src2Mask    (OpMask << Src2Shift)
162 #define Mmx         ((u64)1 << 40)  /* MMX Vector instruction */
163 #define AlignMask   ((u64)7 << 41)
164 #define Aligned     ((u64)1 << 41)  /* Explicitly aligned (e.g. MOVDQA) */
165 #define Unaligned   ((u64)2 << 41)  /* Explicitly unaligned (e.g. MOVDQU) */
166 #define Avx         ((u64)3 << 41)  /* Advanced Vector Extensions */
167 #define Aligned16   ((u64)4 << 41)  /* Aligned to 16 byte boundary (e.g. FXSAVE) */
168 #define Fastop      ((u64)1 << 44)  /* Use opcode::u.fastop */
169 #define NoWrite     ((u64)1 << 45)  /* No writeback */
170 #define SrcWrite    ((u64)1 << 46)  /* Write back src operand */
171 #define NoMod	    ((u64)1 << 47)  /* Mod field is ignored */
172 #define Intercept   ((u64)1 << 48)  /* Has valid intercept field */
173 #define CheckPerm   ((u64)1 << 49)  /* Has valid check_perm field */
174 #define PrivUD      ((u64)1 << 51)  /* #UD instead of #GP on CPL > 0 */
175 #define NearBranch  ((u64)1 << 52)  /* Near branches */
176 #define No16	    ((u64)1 << 53)  /* No 16 bit operand */
177 #define IncSP       ((u64)1 << 54)  /* SP is incremented before ModRM calc */
178 #define TwoMemOp    ((u64)1 << 55)  /* Instruction has two memory operand */
179 #define IsBranch    ((u64)1 << 56)  /* Instruction is considered a branch. */
180 
181 #define DstXacc     (DstAccLo | SrcAccHi | SrcWrite)
182 
183 #define X2(x...) x, x
184 #define X3(x...) X2(x), x
185 #define X4(x...) X2(x), X2(x)
186 #define X5(x...) X4(x), x
187 #define X6(x...) X4(x), X2(x)
188 #define X7(x...) X4(x), X3(x)
189 #define X8(x...) X4(x), X4(x)
190 #define X16(x...) X8(x), X8(x)
191 
192 struct opcode {
193 	u64 flags;
194 	u8 intercept;
195 	u8 pad[7];
196 	union {
197 		int (*execute)(struct x86_emulate_ctxt *ctxt);
198 		const struct opcode *group;
199 		const struct group_dual *gdual;
200 		const struct gprefix *gprefix;
201 		const struct escape *esc;
202 		const struct instr_dual *idual;
203 		const struct mode_dual *mdual;
204 		void (*fastop)(struct fastop *fake);
205 	} u;
206 	int (*check_perm)(struct x86_emulate_ctxt *ctxt);
207 };
208 
209 struct group_dual {
210 	struct opcode mod012[8];
211 	struct opcode mod3[8];
212 };
213 
214 struct gprefix {
215 	struct opcode pfx_no;
216 	struct opcode pfx_66;
217 	struct opcode pfx_f2;
218 	struct opcode pfx_f3;
219 };
220 
221 struct escape {
222 	struct opcode op[8];
223 	struct opcode high[64];
224 };
225 
226 struct instr_dual {
227 	struct opcode mod012;
228 	struct opcode mod3;
229 };
230 
231 struct mode_dual {
232 	struct opcode mode32;
233 	struct opcode mode64;
234 };
235 
236 #define EFLG_RESERVED_ZEROS_MASK 0xffc0802a
237 
238 enum x86_transfer_type {
239 	X86_TRANSFER_NONE,
240 	X86_TRANSFER_CALL_JMP,
241 	X86_TRANSFER_RET,
242 	X86_TRANSFER_TASK_SWITCH,
243 };
244 
245 static ulong reg_read(struct x86_emulate_ctxt *ctxt, unsigned nr)
246 {
247 	if (KVM_EMULATOR_BUG_ON(nr >= NR_EMULATOR_GPRS, ctxt))
248 		nr &= NR_EMULATOR_GPRS - 1;
249 
250 	if (!(ctxt->regs_valid & (1 << nr))) {
251 		ctxt->regs_valid |= 1 << nr;
252 		ctxt->_regs[nr] = ctxt->ops->read_gpr(ctxt, nr);
253 	}
254 	return ctxt->_regs[nr];
255 }
256 
257 static ulong *reg_write(struct x86_emulate_ctxt *ctxt, unsigned nr)
258 {
259 	if (KVM_EMULATOR_BUG_ON(nr >= NR_EMULATOR_GPRS, ctxt))
260 		nr &= NR_EMULATOR_GPRS - 1;
261 
262 	BUILD_BUG_ON(sizeof(ctxt->regs_dirty) * BITS_PER_BYTE < NR_EMULATOR_GPRS);
263 	BUILD_BUG_ON(sizeof(ctxt->regs_valid) * BITS_PER_BYTE < NR_EMULATOR_GPRS);
264 
265 	ctxt->regs_valid |= 1 << nr;
266 	ctxt->regs_dirty |= 1 << nr;
267 	return &ctxt->_regs[nr];
268 }
269 
270 static ulong *reg_rmw(struct x86_emulate_ctxt *ctxt, unsigned nr)
271 {
272 	reg_read(ctxt, nr);
273 	return reg_write(ctxt, nr);
274 }
275 
276 static void writeback_registers(struct x86_emulate_ctxt *ctxt)
277 {
278 	unsigned long dirty = ctxt->regs_dirty;
279 	unsigned reg;
280 
281 	for_each_set_bit(reg, &dirty, NR_EMULATOR_GPRS)
282 		ctxt->ops->write_gpr(ctxt, reg, ctxt->_regs[reg]);
283 }
284 
285 static void invalidate_registers(struct x86_emulate_ctxt *ctxt)
286 {
287 	ctxt->regs_dirty = 0;
288 	ctxt->regs_valid = 0;
289 }
290 
291 /*
292  * These EFLAGS bits are restored from saved value during emulation, and
293  * any changes are written back to the saved value after emulation.
294  */
295 #define EFLAGS_MASK (X86_EFLAGS_OF|X86_EFLAGS_SF|X86_EFLAGS_ZF|X86_EFLAGS_AF|\
296 		     X86_EFLAGS_PF|X86_EFLAGS_CF)
297 
298 #ifdef CONFIG_X86_64
299 #define ON64(x) x
300 #else
301 #define ON64(x)
302 #endif
303 
304 /*
305  * fastop functions have a special calling convention:
306  *
307  * dst:    rax        (in/out)
308  * src:    rdx        (in/out)
309  * src2:   rcx        (in)
310  * flags:  rflags     (in/out)
311  * ex:     rsi        (in:fastop pointer, out:zero if exception)
312  *
313  * Moreover, they are all exactly FASTOP_SIZE bytes long, so functions for
314  * different operand sizes can be reached by calculation, rather than a jump
315  * table (which would be bigger than the code).
316  *
317  * The 16 byte alignment, considering 5 bytes for the RET thunk, 3 for ENDBR
318  * and 1 for the straight line speculation INT3, leaves 7 bytes for the
319  * body of the function.  Currently none is larger than 4.
320  */
321 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop);
322 
323 #define FASTOP_SIZE	16
324 
325 #define __FOP_FUNC(name) \
326 	".align " __stringify(FASTOP_SIZE) " \n\t" \
327 	".type " name ", @function \n\t" \
328 	name ":\n\t" \
329 	ASM_ENDBR
330 
331 #define FOP_FUNC(name) \
332 	__FOP_FUNC(#name)
333 
334 #define __FOP_RET(name) \
335 	"11: " ASM_RET \
336 	".size " name ", .-" name "\n\t"
337 
338 #define FOP_RET(name) \
339 	__FOP_RET(#name)
340 
341 #define __FOP_START(op, align) \
342 	extern void em_##op(struct fastop *fake); \
343 	asm(".pushsection .text, \"ax\" \n\t" \
344 	    ".global em_" #op " \n\t" \
345 	    ".align " __stringify(align) " \n\t" \
346 	    "em_" #op ":\n\t"
347 
348 #define FOP_START(op) __FOP_START(op, FASTOP_SIZE)
349 
350 #define FOP_END \
351 	    ".popsection")
352 
353 #define __FOPNOP(name) \
354 	__FOP_FUNC(name) \
355 	__FOP_RET(name)
356 
357 #define FOPNOP() \
358 	__FOPNOP(__stringify(__UNIQUE_ID(nop)))
359 
360 #define FOP1E(op,  dst) \
361 	__FOP_FUNC(#op "_" #dst) \
362 	"10: " #op " %" #dst " \n\t" \
363 	__FOP_RET(#op "_" #dst)
364 
365 #define FOP1EEX(op,  dst) \
366 	FOP1E(op, dst) _ASM_EXTABLE_TYPE_REG(10b, 11b, EX_TYPE_ZERO_REG, %%esi)
367 
368 #define FASTOP1(op) \
369 	FOP_START(op) \
370 	FOP1E(op##b, al) \
371 	FOP1E(op##w, ax) \
372 	FOP1E(op##l, eax) \
373 	ON64(FOP1E(op##q, rax))	\
374 	FOP_END
375 
376 /* 1-operand, using src2 (for MUL/DIV r/m) */
377 #define FASTOP1SRC2(op, name) \
378 	FOP_START(name) \
379 	FOP1E(op, cl) \
380 	FOP1E(op, cx) \
381 	FOP1E(op, ecx) \
382 	ON64(FOP1E(op, rcx)) \
383 	FOP_END
384 
385 /* 1-operand, using src2 (for MUL/DIV r/m), with exceptions */
386 #define FASTOP1SRC2EX(op, name) \
387 	FOP_START(name) \
388 	FOP1EEX(op, cl) \
389 	FOP1EEX(op, cx) \
390 	FOP1EEX(op, ecx) \
391 	ON64(FOP1EEX(op, rcx)) \
392 	FOP_END
393 
394 #define FOP2E(op,  dst, src)	   \
395 	__FOP_FUNC(#op "_" #dst "_" #src) \
396 	#op " %" #src ", %" #dst " \n\t" \
397 	__FOP_RET(#op "_" #dst "_" #src)
398 
399 #define FASTOP2(op) \
400 	FOP_START(op) \
401 	FOP2E(op##b, al, dl) \
402 	FOP2E(op##w, ax, dx) \
403 	FOP2E(op##l, eax, edx) \
404 	ON64(FOP2E(op##q, rax, rdx)) \
405 	FOP_END
406 
407 /* 2 operand, word only */
408 #define FASTOP2W(op) \
409 	FOP_START(op) \
410 	FOPNOP() \
411 	FOP2E(op##w, ax, dx) \
412 	FOP2E(op##l, eax, edx) \
413 	ON64(FOP2E(op##q, rax, rdx)) \
414 	FOP_END
415 
416 /* 2 operand, src is CL */
417 #define FASTOP2CL(op) \
418 	FOP_START(op) \
419 	FOP2E(op##b, al, cl) \
420 	FOP2E(op##w, ax, cl) \
421 	FOP2E(op##l, eax, cl) \
422 	ON64(FOP2E(op##q, rax, cl)) \
423 	FOP_END
424 
425 /* 2 operand, src and dest are reversed */
426 #define FASTOP2R(op, name) \
427 	FOP_START(name) \
428 	FOP2E(op##b, dl, al) \
429 	FOP2E(op##w, dx, ax) \
430 	FOP2E(op##l, edx, eax) \
431 	ON64(FOP2E(op##q, rdx, rax)) \
432 	FOP_END
433 
434 #define FOP3E(op,  dst, src, src2) \
435 	__FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \
436 	#op " %" #src2 ", %" #src ", %" #dst " \n\t"\
437 	__FOP_RET(#op "_" #dst "_" #src "_" #src2)
438 
439 /* 3-operand, word-only, src2=cl */
440 #define FASTOP3WCL(op) \
441 	FOP_START(op) \
442 	FOPNOP() \
443 	FOP3E(op##w, ax, dx, cl) \
444 	FOP3E(op##l, eax, edx, cl) \
445 	ON64(FOP3E(op##q, rax, rdx, cl)) \
446 	FOP_END
447 
448 /* Special case for SETcc - 1 instruction per cc */
449 
450 /*
451  * Depending on .config the SETcc functions look like:
452  *
453  * ENDBR			[4 bytes; CONFIG_X86_KERNEL_IBT]
454  * SETcc %al			[3 bytes]
455  * RET | JMP __x86_return_thunk	[1,5 bytes; CONFIG_RETHUNK]
456  * INT3				[1 byte; CONFIG_SLS]
457  */
458 #define SETCC_ALIGN	16
459 
460 #define FOP_SETCC(op) \
461 	".align " __stringify(SETCC_ALIGN) " \n\t" \
462 	".type " #op ", @function \n\t" \
463 	#op ": \n\t" \
464 	ASM_ENDBR \
465 	#op " %al \n\t" \
466 	__FOP_RET(#op) \
467 	".skip " __stringify(SETCC_ALIGN) " - (.-" #op "), 0xcc \n\t"
468 
469 __FOP_START(setcc, SETCC_ALIGN)
470 FOP_SETCC(seto)
471 FOP_SETCC(setno)
472 FOP_SETCC(setc)
473 FOP_SETCC(setnc)
474 FOP_SETCC(setz)
475 FOP_SETCC(setnz)
476 FOP_SETCC(setbe)
477 FOP_SETCC(setnbe)
478 FOP_SETCC(sets)
479 FOP_SETCC(setns)
480 FOP_SETCC(setp)
481 FOP_SETCC(setnp)
482 FOP_SETCC(setl)
483 FOP_SETCC(setnl)
484 FOP_SETCC(setle)
485 FOP_SETCC(setnle)
486 FOP_END;
487 
488 FOP_START(salc)
489 FOP_FUNC(salc)
490 "pushf; sbb %al, %al; popf \n\t"
491 FOP_RET(salc)
492 FOP_END;
493 
494 /*
495  * XXX: inoutclob user must know where the argument is being expanded.
496  *      Relying on CONFIG_CC_HAS_ASM_GOTO would allow us to remove _fault.
497  */
498 #define asm_safe(insn, inoutclob...) \
499 ({ \
500 	int _fault = 0; \
501  \
502 	asm volatile("1:" insn "\n" \
503 	             "2:\n" \
504 		     _ASM_EXTABLE_TYPE_REG(1b, 2b, EX_TYPE_ONE_REG, %[_fault]) \
505 	             : [_fault] "+r"(_fault) inoutclob ); \
506  \
507 	_fault ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE; \
508 })
509 
510 static int emulator_check_intercept(struct x86_emulate_ctxt *ctxt,
511 				    enum x86_intercept intercept,
512 				    enum x86_intercept_stage stage)
513 {
514 	struct x86_instruction_info info = {
515 		.intercept  = intercept,
516 		.rep_prefix = ctxt->rep_prefix,
517 		.modrm_mod  = ctxt->modrm_mod,
518 		.modrm_reg  = ctxt->modrm_reg,
519 		.modrm_rm   = ctxt->modrm_rm,
520 		.src_val    = ctxt->src.val64,
521 		.dst_val    = ctxt->dst.val64,
522 		.src_bytes  = ctxt->src.bytes,
523 		.dst_bytes  = ctxt->dst.bytes,
524 		.ad_bytes   = ctxt->ad_bytes,
525 		.next_rip   = ctxt->eip,
526 	};
527 
528 	return ctxt->ops->intercept(ctxt, &info, stage);
529 }
530 
531 static void assign_masked(ulong *dest, ulong src, ulong mask)
532 {
533 	*dest = (*dest & ~mask) | (src & mask);
534 }
535 
536 static void assign_register(unsigned long *reg, u64 val, int bytes)
537 {
538 	/* The 4-byte case *is* correct: in 64-bit mode we zero-extend. */
539 	switch (bytes) {
540 	case 1:
541 		*(u8 *)reg = (u8)val;
542 		break;
543 	case 2:
544 		*(u16 *)reg = (u16)val;
545 		break;
546 	case 4:
547 		*reg = (u32)val;
548 		break;	/* 64b: zero-extend */
549 	case 8:
550 		*reg = val;
551 		break;
552 	}
553 }
554 
555 static inline unsigned long ad_mask(struct x86_emulate_ctxt *ctxt)
556 {
557 	return (1UL << (ctxt->ad_bytes << 3)) - 1;
558 }
559 
560 static ulong stack_mask(struct x86_emulate_ctxt *ctxt)
561 {
562 	u16 sel;
563 	struct desc_struct ss;
564 
565 	if (ctxt->mode == X86EMUL_MODE_PROT64)
566 		return ~0UL;
567 	ctxt->ops->get_segment(ctxt, &sel, &ss, NULL, VCPU_SREG_SS);
568 	return ~0U >> ((ss.d ^ 1) * 16);  /* d=0: 0xffff; d=1: 0xffffffff */
569 }
570 
571 static int stack_size(struct x86_emulate_ctxt *ctxt)
572 {
573 	return (__fls(stack_mask(ctxt)) + 1) >> 3;
574 }
575 
576 /* Access/update address held in a register, based on addressing mode. */
577 static inline unsigned long
578 address_mask(struct x86_emulate_ctxt *ctxt, unsigned long reg)
579 {
580 	if (ctxt->ad_bytes == sizeof(unsigned long))
581 		return reg;
582 	else
583 		return reg & ad_mask(ctxt);
584 }
585 
586 static inline unsigned long
587 register_address(struct x86_emulate_ctxt *ctxt, int reg)
588 {
589 	return address_mask(ctxt, reg_read(ctxt, reg));
590 }
591 
592 static void masked_increment(ulong *reg, ulong mask, int inc)
593 {
594 	assign_masked(reg, *reg + inc, mask);
595 }
596 
597 static inline void
598 register_address_increment(struct x86_emulate_ctxt *ctxt, int reg, int inc)
599 {
600 	ulong *preg = reg_rmw(ctxt, reg);
601 
602 	assign_register(preg, *preg + inc, ctxt->ad_bytes);
603 }
604 
605 static void rsp_increment(struct x86_emulate_ctxt *ctxt, int inc)
606 {
607 	masked_increment(reg_rmw(ctxt, VCPU_REGS_RSP), stack_mask(ctxt), inc);
608 }
609 
610 static u32 desc_limit_scaled(struct desc_struct *desc)
611 {
612 	u32 limit = get_desc_limit(desc);
613 
614 	return desc->g ? (limit << 12) | 0xfff : limit;
615 }
616 
617 static unsigned long seg_base(struct x86_emulate_ctxt *ctxt, int seg)
618 {
619 	if (ctxt->mode == X86EMUL_MODE_PROT64 && seg < VCPU_SREG_FS)
620 		return 0;
621 
622 	return ctxt->ops->get_cached_segment_base(ctxt, seg);
623 }
624 
625 static int emulate_exception(struct x86_emulate_ctxt *ctxt, int vec,
626 			     u32 error, bool valid)
627 {
628 	if (KVM_EMULATOR_BUG_ON(vec > 0x1f, ctxt))
629 		return X86EMUL_UNHANDLEABLE;
630 
631 	ctxt->exception.vector = vec;
632 	ctxt->exception.error_code = error;
633 	ctxt->exception.error_code_valid = valid;
634 	return X86EMUL_PROPAGATE_FAULT;
635 }
636 
637 static int emulate_db(struct x86_emulate_ctxt *ctxt)
638 {
639 	return emulate_exception(ctxt, DB_VECTOR, 0, false);
640 }
641 
642 static int emulate_gp(struct x86_emulate_ctxt *ctxt, int err)
643 {
644 	return emulate_exception(ctxt, GP_VECTOR, err, true);
645 }
646 
647 static int emulate_ss(struct x86_emulate_ctxt *ctxt, int err)
648 {
649 	return emulate_exception(ctxt, SS_VECTOR, err, true);
650 }
651 
652 static int emulate_ud(struct x86_emulate_ctxt *ctxt)
653 {
654 	return emulate_exception(ctxt, UD_VECTOR, 0, false);
655 }
656 
657 static int emulate_ts(struct x86_emulate_ctxt *ctxt, int err)
658 {
659 	return emulate_exception(ctxt, TS_VECTOR, err, true);
660 }
661 
662 static int emulate_de(struct x86_emulate_ctxt *ctxt)
663 {
664 	return emulate_exception(ctxt, DE_VECTOR, 0, false);
665 }
666 
667 static int emulate_nm(struct x86_emulate_ctxt *ctxt)
668 {
669 	return emulate_exception(ctxt, NM_VECTOR, 0, false);
670 }
671 
672 static u16 get_segment_selector(struct x86_emulate_ctxt *ctxt, unsigned seg)
673 {
674 	u16 selector;
675 	struct desc_struct desc;
676 
677 	ctxt->ops->get_segment(ctxt, &selector, &desc, NULL, seg);
678 	return selector;
679 }
680 
681 static void set_segment_selector(struct x86_emulate_ctxt *ctxt, u16 selector,
682 				 unsigned seg)
683 {
684 	u16 dummy;
685 	u32 base3;
686 	struct desc_struct desc;
687 
688 	ctxt->ops->get_segment(ctxt, &dummy, &desc, &base3, seg);
689 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, seg);
690 }
691 
692 static inline u8 ctxt_virt_addr_bits(struct x86_emulate_ctxt *ctxt)
693 {
694 	return (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_LA57) ? 57 : 48;
695 }
696 
697 static inline bool emul_is_noncanonical_address(u64 la,
698 						struct x86_emulate_ctxt *ctxt)
699 {
700 	return !__is_canonical_address(la, ctxt_virt_addr_bits(ctxt));
701 }
702 
703 /*
704  * x86 defines three classes of vector instructions: explicitly
705  * aligned, explicitly unaligned, and the rest, which change behaviour
706  * depending on whether they're AVX encoded or not.
707  *
708  * Also included is CMPXCHG16B which is not a vector instruction, yet it is
709  * subject to the same check.  FXSAVE and FXRSTOR are checked here too as their
710  * 512 bytes of data must be aligned to a 16 byte boundary.
711  */
712 static unsigned insn_alignment(struct x86_emulate_ctxt *ctxt, unsigned size)
713 {
714 	u64 alignment = ctxt->d & AlignMask;
715 
716 	if (likely(size < 16))
717 		return 1;
718 
719 	switch (alignment) {
720 	case Unaligned:
721 	case Avx:
722 		return 1;
723 	case Aligned16:
724 		return 16;
725 	case Aligned:
726 	default:
727 		return size;
728 	}
729 }
730 
731 static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt,
732 				       struct segmented_address addr,
733 				       unsigned *max_size, unsigned size,
734 				       bool write, bool fetch,
735 				       enum x86emul_mode mode, ulong *linear)
736 {
737 	struct desc_struct desc;
738 	bool usable;
739 	ulong la;
740 	u32 lim;
741 	u16 sel;
742 	u8  va_bits;
743 
744 	la = seg_base(ctxt, addr.seg) + addr.ea;
745 	*max_size = 0;
746 	switch (mode) {
747 	case X86EMUL_MODE_PROT64:
748 		*linear = la;
749 		va_bits = ctxt_virt_addr_bits(ctxt);
750 		if (!__is_canonical_address(la, va_bits))
751 			goto bad;
752 
753 		*max_size = min_t(u64, ~0u, (1ull << va_bits) - la);
754 		if (size > *max_size)
755 			goto bad;
756 		break;
757 	default:
758 		*linear = la = (u32)la;
759 		usable = ctxt->ops->get_segment(ctxt, &sel, &desc, NULL,
760 						addr.seg);
761 		if (!usable)
762 			goto bad;
763 		/* code segment in protected mode or read-only data segment */
764 		if ((((ctxt->mode != X86EMUL_MODE_REAL) && (desc.type & 8))
765 					|| !(desc.type & 2)) && write)
766 			goto bad;
767 		/* unreadable code segment */
768 		if (!fetch && (desc.type & 8) && !(desc.type & 2))
769 			goto bad;
770 		lim = desc_limit_scaled(&desc);
771 		if (!(desc.type & 8) && (desc.type & 4)) {
772 			/* expand-down segment */
773 			if (addr.ea <= lim)
774 				goto bad;
775 			lim = desc.d ? 0xffffffff : 0xffff;
776 		}
777 		if (addr.ea > lim)
778 			goto bad;
779 		if (lim == 0xffffffff)
780 			*max_size = ~0u;
781 		else {
782 			*max_size = (u64)lim + 1 - addr.ea;
783 			if (size > *max_size)
784 				goto bad;
785 		}
786 		break;
787 	}
788 	if (la & (insn_alignment(ctxt, size) - 1))
789 		return emulate_gp(ctxt, 0);
790 	return X86EMUL_CONTINUE;
791 bad:
792 	if (addr.seg == VCPU_SREG_SS)
793 		return emulate_ss(ctxt, 0);
794 	else
795 		return emulate_gp(ctxt, 0);
796 }
797 
798 static int linearize(struct x86_emulate_ctxt *ctxt,
799 		     struct segmented_address addr,
800 		     unsigned size, bool write,
801 		     ulong *linear)
802 {
803 	unsigned max_size;
804 	return __linearize(ctxt, addr, &max_size, size, write, false,
805 			   ctxt->mode, linear);
806 }
807 
808 static inline int assign_eip(struct x86_emulate_ctxt *ctxt, ulong dst,
809 			     enum x86emul_mode mode)
810 {
811 	ulong linear;
812 	int rc;
813 	unsigned max_size;
814 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
815 					   .ea = dst };
816 
817 	if (ctxt->op_bytes != sizeof(unsigned long))
818 		addr.ea = dst & ((1UL << (ctxt->op_bytes << 3)) - 1);
819 	rc = __linearize(ctxt, addr, &max_size, 1, false, true, mode, &linear);
820 	if (rc == X86EMUL_CONTINUE)
821 		ctxt->_eip = addr.ea;
822 	return rc;
823 }
824 
825 static inline int assign_eip_near(struct x86_emulate_ctxt *ctxt, ulong dst)
826 {
827 	return assign_eip(ctxt, dst, ctxt->mode);
828 }
829 
830 static int assign_eip_far(struct x86_emulate_ctxt *ctxt, ulong dst,
831 			  const struct desc_struct *cs_desc)
832 {
833 	enum x86emul_mode mode = ctxt->mode;
834 	int rc;
835 
836 #ifdef CONFIG_X86_64
837 	if (ctxt->mode >= X86EMUL_MODE_PROT16) {
838 		if (cs_desc->l) {
839 			u64 efer = 0;
840 
841 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
842 			if (efer & EFER_LMA)
843 				mode = X86EMUL_MODE_PROT64;
844 		} else
845 			mode = X86EMUL_MODE_PROT32; /* temporary value */
846 	}
847 #endif
848 	if (mode == X86EMUL_MODE_PROT16 || mode == X86EMUL_MODE_PROT32)
849 		mode = cs_desc->d ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
850 	rc = assign_eip(ctxt, dst, mode);
851 	if (rc == X86EMUL_CONTINUE)
852 		ctxt->mode = mode;
853 	return rc;
854 }
855 
856 static inline int jmp_rel(struct x86_emulate_ctxt *ctxt, int rel)
857 {
858 	return assign_eip_near(ctxt, ctxt->_eip + rel);
859 }
860 
861 static int linear_read_system(struct x86_emulate_ctxt *ctxt, ulong linear,
862 			      void *data, unsigned size)
863 {
864 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, true);
865 }
866 
867 static int linear_write_system(struct x86_emulate_ctxt *ctxt,
868 			       ulong linear, void *data,
869 			       unsigned int size)
870 {
871 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, true);
872 }
873 
874 static int segmented_read_std(struct x86_emulate_ctxt *ctxt,
875 			      struct segmented_address addr,
876 			      void *data,
877 			      unsigned size)
878 {
879 	int rc;
880 	ulong linear;
881 
882 	rc = linearize(ctxt, addr, size, false, &linear);
883 	if (rc != X86EMUL_CONTINUE)
884 		return rc;
885 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, false);
886 }
887 
888 static int segmented_write_std(struct x86_emulate_ctxt *ctxt,
889 			       struct segmented_address addr,
890 			       void *data,
891 			       unsigned int size)
892 {
893 	int rc;
894 	ulong linear;
895 
896 	rc = linearize(ctxt, addr, size, true, &linear);
897 	if (rc != X86EMUL_CONTINUE)
898 		return rc;
899 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, false);
900 }
901 
902 /*
903  * Prefetch the remaining bytes of the instruction without crossing page
904  * boundary if they are not in fetch_cache yet.
905  */
906 static int __do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt, int op_size)
907 {
908 	int rc;
909 	unsigned size, max_size;
910 	unsigned long linear;
911 	int cur_size = ctxt->fetch.end - ctxt->fetch.data;
912 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
913 					   .ea = ctxt->eip + cur_size };
914 
915 	/*
916 	 * We do not know exactly how many bytes will be needed, and
917 	 * __linearize is expensive, so fetch as much as possible.  We
918 	 * just have to avoid going beyond the 15 byte limit, the end
919 	 * of the segment, or the end of the page.
920 	 *
921 	 * __linearize is called with size 0 so that it does not do any
922 	 * boundary check itself.  Instead, we use max_size to check
923 	 * against op_size.
924 	 */
925 	rc = __linearize(ctxt, addr, &max_size, 0, false, true, ctxt->mode,
926 			 &linear);
927 	if (unlikely(rc != X86EMUL_CONTINUE))
928 		return rc;
929 
930 	size = min_t(unsigned, 15UL ^ cur_size, max_size);
931 	size = min_t(unsigned, size, PAGE_SIZE - offset_in_page(linear));
932 
933 	/*
934 	 * One instruction can only straddle two pages,
935 	 * and one has been loaded at the beginning of
936 	 * x86_decode_insn.  So, if not enough bytes
937 	 * still, we must have hit the 15-byte boundary.
938 	 */
939 	if (unlikely(size < op_size))
940 		return emulate_gp(ctxt, 0);
941 
942 	rc = ctxt->ops->fetch(ctxt, linear, ctxt->fetch.end,
943 			      size, &ctxt->exception);
944 	if (unlikely(rc != X86EMUL_CONTINUE))
945 		return rc;
946 	ctxt->fetch.end += size;
947 	return X86EMUL_CONTINUE;
948 }
949 
950 static __always_inline int do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt,
951 					       unsigned size)
952 {
953 	unsigned done_size = ctxt->fetch.end - ctxt->fetch.ptr;
954 
955 	if (unlikely(done_size < size))
956 		return __do_insn_fetch_bytes(ctxt, size - done_size);
957 	else
958 		return X86EMUL_CONTINUE;
959 }
960 
961 /* Fetch next part of the instruction being emulated. */
962 #define insn_fetch(_type, _ctxt)					\
963 ({	_type _x;							\
964 									\
965 	rc = do_insn_fetch_bytes(_ctxt, sizeof(_type));			\
966 	if (rc != X86EMUL_CONTINUE)					\
967 		goto done;						\
968 	ctxt->_eip += sizeof(_type);					\
969 	memcpy(&_x, ctxt->fetch.ptr, sizeof(_type));			\
970 	ctxt->fetch.ptr += sizeof(_type);				\
971 	_x;								\
972 })
973 
974 #define insn_fetch_arr(_arr, _size, _ctxt)				\
975 ({									\
976 	rc = do_insn_fetch_bytes(_ctxt, _size);				\
977 	if (rc != X86EMUL_CONTINUE)					\
978 		goto done;						\
979 	ctxt->_eip += (_size);						\
980 	memcpy(_arr, ctxt->fetch.ptr, _size);				\
981 	ctxt->fetch.ptr += (_size);					\
982 })
983 
984 /*
985  * Given the 'reg' portion of a ModRM byte, and a register block, return a
986  * pointer into the block that addresses the relevant register.
987  * @highbyte_regs specifies whether to decode AH,CH,DH,BH.
988  */
989 static void *decode_register(struct x86_emulate_ctxt *ctxt, u8 modrm_reg,
990 			     int byteop)
991 {
992 	void *p;
993 	int highbyte_regs = (ctxt->rex_prefix == 0) && byteop;
994 
995 	if (highbyte_regs && modrm_reg >= 4 && modrm_reg < 8)
996 		p = (unsigned char *)reg_rmw(ctxt, modrm_reg & 3) + 1;
997 	else
998 		p = reg_rmw(ctxt, modrm_reg);
999 	return p;
1000 }
1001 
1002 static int read_descriptor(struct x86_emulate_ctxt *ctxt,
1003 			   struct segmented_address addr,
1004 			   u16 *size, unsigned long *address, int op_bytes)
1005 {
1006 	int rc;
1007 
1008 	if (op_bytes == 2)
1009 		op_bytes = 3;
1010 	*address = 0;
1011 	rc = segmented_read_std(ctxt, addr, size, 2);
1012 	if (rc != X86EMUL_CONTINUE)
1013 		return rc;
1014 	addr.ea += 2;
1015 	rc = segmented_read_std(ctxt, addr, address, op_bytes);
1016 	return rc;
1017 }
1018 
1019 FASTOP2(add);
1020 FASTOP2(or);
1021 FASTOP2(adc);
1022 FASTOP2(sbb);
1023 FASTOP2(and);
1024 FASTOP2(sub);
1025 FASTOP2(xor);
1026 FASTOP2(cmp);
1027 FASTOP2(test);
1028 
1029 FASTOP1SRC2(mul, mul_ex);
1030 FASTOP1SRC2(imul, imul_ex);
1031 FASTOP1SRC2EX(div, div_ex);
1032 FASTOP1SRC2EX(idiv, idiv_ex);
1033 
1034 FASTOP3WCL(shld);
1035 FASTOP3WCL(shrd);
1036 
1037 FASTOP2W(imul);
1038 
1039 FASTOP1(not);
1040 FASTOP1(neg);
1041 FASTOP1(inc);
1042 FASTOP1(dec);
1043 
1044 FASTOP2CL(rol);
1045 FASTOP2CL(ror);
1046 FASTOP2CL(rcl);
1047 FASTOP2CL(rcr);
1048 FASTOP2CL(shl);
1049 FASTOP2CL(shr);
1050 FASTOP2CL(sar);
1051 
1052 FASTOP2W(bsf);
1053 FASTOP2W(bsr);
1054 FASTOP2W(bt);
1055 FASTOP2W(bts);
1056 FASTOP2W(btr);
1057 FASTOP2W(btc);
1058 
1059 FASTOP2(xadd);
1060 
1061 FASTOP2R(cmp, cmp_r);
1062 
1063 static int em_bsf_c(struct x86_emulate_ctxt *ctxt)
1064 {
1065 	/* If src is zero, do not writeback, but update flags */
1066 	if (ctxt->src.val == 0)
1067 		ctxt->dst.type = OP_NONE;
1068 	return fastop(ctxt, em_bsf);
1069 }
1070 
1071 static int em_bsr_c(struct x86_emulate_ctxt *ctxt)
1072 {
1073 	/* If src is zero, do not writeback, but update flags */
1074 	if (ctxt->src.val == 0)
1075 		ctxt->dst.type = OP_NONE;
1076 	return fastop(ctxt, em_bsr);
1077 }
1078 
1079 static __always_inline u8 test_cc(unsigned int condition, unsigned long flags)
1080 {
1081 	u8 rc;
1082 	void (*fop)(void) = (void *)em_setcc + SETCC_ALIGN * (condition & 0xf);
1083 
1084 	flags = (flags & EFLAGS_MASK) | X86_EFLAGS_IF;
1085 	asm("push %[flags]; popf; " CALL_NOSPEC
1086 	    : "=a"(rc) : [thunk_target]"r"(fop), [flags]"r"(flags));
1087 	return rc;
1088 }
1089 
1090 static void fetch_register_operand(struct operand *op)
1091 {
1092 	switch (op->bytes) {
1093 	case 1:
1094 		op->val = *(u8 *)op->addr.reg;
1095 		break;
1096 	case 2:
1097 		op->val = *(u16 *)op->addr.reg;
1098 		break;
1099 	case 4:
1100 		op->val = *(u32 *)op->addr.reg;
1101 		break;
1102 	case 8:
1103 		op->val = *(u64 *)op->addr.reg;
1104 		break;
1105 	}
1106 }
1107 
1108 static int em_fninit(struct x86_emulate_ctxt *ctxt)
1109 {
1110 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1111 		return emulate_nm(ctxt);
1112 
1113 	kvm_fpu_get();
1114 	asm volatile("fninit");
1115 	kvm_fpu_put();
1116 	return X86EMUL_CONTINUE;
1117 }
1118 
1119 static int em_fnstcw(struct x86_emulate_ctxt *ctxt)
1120 {
1121 	u16 fcw;
1122 
1123 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1124 		return emulate_nm(ctxt);
1125 
1126 	kvm_fpu_get();
1127 	asm volatile("fnstcw %0": "+m"(fcw));
1128 	kvm_fpu_put();
1129 
1130 	ctxt->dst.val = fcw;
1131 
1132 	return X86EMUL_CONTINUE;
1133 }
1134 
1135 static int em_fnstsw(struct x86_emulate_ctxt *ctxt)
1136 {
1137 	u16 fsw;
1138 
1139 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1140 		return emulate_nm(ctxt);
1141 
1142 	kvm_fpu_get();
1143 	asm volatile("fnstsw %0": "+m"(fsw));
1144 	kvm_fpu_put();
1145 
1146 	ctxt->dst.val = fsw;
1147 
1148 	return X86EMUL_CONTINUE;
1149 }
1150 
1151 static void decode_register_operand(struct x86_emulate_ctxt *ctxt,
1152 				    struct operand *op)
1153 {
1154 	unsigned reg = ctxt->modrm_reg;
1155 
1156 	if (!(ctxt->d & ModRM))
1157 		reg = (ctxt->b & 7) | ((ctxt->rex_prefix & 1) << 3);
1158 
1159 	if (ctxt->d & Sse) {
1160 		op->type = OP_XMM;
1161 		op->bytes = 16;
1162 		op->addr.xmm = reg;
1163 		kvm_read_sse_reg(reg, &op->vec_val);
1164 		return;
1165 	}
1166 	if (ctxt->d & Mmx) {
1167 		reg &= 7;
1168 		op->type = OP_MM;
1169 		op->bytes = 8;
1170 		op->addr.mm = reg;
1171 		return;
1172 	}
1173 
1174 	op->type = OP_REG;
1175 	op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1176 	op->addr.reg = decode_register(ctxt, reg, ctxt->d & ByteOp);
1177 
1178 	fetch_register_operand(op);
1179 	op->orig_val = op->val;
1180 }
1181 
1182 static void adjust_modrm_seg(struct x86_emulate_ctxt *ctxt, int base_reg)
1183 {
1184 	if (base_reg == VCPU_REGS_RSP || base_reg == VCPU_REGS_RBP)
1185 		ctxt->modrm_seg = VCPU_SREG_SS;
1186 }
1187 
1188 static int decode_modrm(struct x86_emulate_ctxt *ctxt,
1189 			struct operand *op)
1190 {
1191 	u8 sib;
1192 	int index_reg, base_reg, scale;
1193 	int rc = X86EMUL_CONTINUE;
1194 	ulong modrm_ea = 0;
1195 
1196 	ctxt->modrm_reg = ((ctxt->rex_prefix << 1) & 8); /* REX.R */
1197 	index_reg = (ctxt->rex_prefix << 2) & 8; /* REX.X */
1198 	base_reg = (ctxt->rex_prefix << 3) & 8; /* REX.B */
1199 
1200 	ctxt->modrm_mod = (ctxt->modrm & 0xc0) >> 6;
1201 	ctxt->modrm_reg |= (ctxt->modrm & 0x38) >> 3;
1202 	ctxt->modrm_rm = base_reg | (ctxt->modrm & 0x07);
1203 	ctxt->modrm_seg = VCPU_SREG_DS;
1204 
1205 	if (ctxt->modrm_mod == 3 || (ctxt->d & NoMod)) {
1206 		op->type = OP_REG;
1207 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1208 		op->addr.reg = decode_register(ctxt, ctxt->modrm_rm,
1209 				ctxt->d & ByteOp);
1210 		if (ctxt->d & Sse) {
1211 			op->type = OP_XMM;
1212 			op->bytes = 16;
1213 			op->addr.xmm = ctxt->modrm_rm;
1214 			kvm_read_sse_reg(ctxt->modrm_rm, &op->vec_val);
1215 			return rc;
1216 		}
1217 		if (ctxt->d & Mmx) {
1218 			op->type = OP_MM;
1219 			op->bytes = 8;
1220 			op->addr.mm = ctxt->modrm_rm & 7;
1221 			return rc;
1222 		}
1223 		fetch_register_operand(op);
1224 		return rc;
1225 	}
1226 
1227 	op->type = OP_MEM;
1228 
1229 	if (ctxt->ad_bytes == 2) {
1230 		unsigned bx = reg_read(ctxt, VCPU_REGS_RBX);
1231 		unsigned bp = reg_read(ctxt, VCPU_REGS_RBP);
1232 		unsigned si = reg_read(ctxt, VCPU_REGS_RSI);
1233 		unsigned di = reg_read(ctxt, VCPU_REGS_RDI);
1234 
1235 		/* 16-bit ModR/M decode. */
1236 		switch (ctxt->modrm_mod) {
1237 		case 0:
1238 			if (ctxt->modrm_rm == 6)
1239 				modrm_ea += insn_fetch(u16, ctxt);
1240 			break;
1241 		case 1:
1242 			modrm_ea += insn_fetch(s8, ctxt);
1243 			break;
1244 		case 2:
1245 			modrm_ea += insn_fetch(u16, ctxt);
1246 			break;
1247 		}
1248 		switch (ctxt->modrm_rm) {
1249 		case 0:
1250 			modrm_ea += bx + si;
1251 			break;
1252 		case 1:
1253 			modrm_ea += bx + di;
1254 			break;
1255 		case 2:
1256 			modrm_ea += bp + si;
1257 			break;
1258 		case 3:
1259 			modrm_ea += bp + di;
1260 			break;
1261 		case 4:
1262 			modrm_ea += si;
1263 			break;
1264 		case 5:
1265 			modrm_ea += di;
1266 			break;
1267 		case 6:
1268 			if (ctxt->modrm_mod != 0)
1269 				modrm_ea += bp;
1270 			break;
1271 		case 7:
1272 			modrm_ea += bx;
1273 			break;
1274 		}
1275 		if (ctxt->modrm_rm == 2 || ctxt->modrm_rm == 3 ||
1276 		    (ctxt->modrm_rm == 6 && ctxt->modrm_mod != 0))
1277 			ctxt->modrm_seg = VCPU_SREG_SS;
1278 		modrm_ea = (u16)modrm_ea;
1279 	} else {
1280 		/* 32/64-bit ModR/M decode. */
1281 		if ((ctxt->modrm_rm & 7) == 4) {
1282 			sib = insn_fetch(u8, ctxt);
1283 			index_reg |= (sib >> 3) & 7;
1284 			base_reg |= sib & 7;
1285 			scale = sib >> 6;
1286 
1287 			if ((base_reg & 7) == 5 && ctxt->modrm_mod == 0)
1288 				modrm_ea += insn_fetch(s32, ctxt);
1289 			else {
1290 				modrm_ea += reg_read(ctxt, base_reg);
1291 				adjust_modrm_seg(ctxt, base_reg);
1292 				/* Increment ESP on POP [ESP] */
1293 				if ((ctxt->d & IncSP) &&
1294 				    base_reg == VCPU_REGS_RSP)
1295 					modrm_ea += ctxt->op_bytes;
1296 			}
1297 			if (index_reg != 4)
1298 				modrm_ea += reg_read(ctxt, index_reg) << scale;
1299 		} else if ((ctxt->modrm_rm & 7) == 5 && ctxt->modrm_mod == 0) {
1300 			modrm_ea += insn_fetch(s32, ctxt);
1301 			if (ctxt->mode == X86EMUL_MODE_PROT64)
1302 				ctxt->rip_relative = 1;
1303 		} else {
1304 			base_reg = ctxt->modrm_rm;
1305 			modrm_ea += reg_read(ctxt, base_reg);
1306 			adjust_modrm_seg(ctxt, base_reg);
1307 		}
1308 		switch (ctxt->modrm_mod) {
1309 		case 1:
1310 			modrm_ea += insn_fetch(s8, ctxt);
1311 			break;
1312 		case 2:
1313 			modrm_ea += insn_fetch(s32, ctxt);
1314 			break;
1315 		}
1316 	}
1317 	op->addr.mem.ea = modrm_ea;
1318 	if (ctxt->ad_bytes != 8)
1319 		ctxt->memop.addr.mem.ea = (u32)ctxt->memop.addr.mem.ea;
1320 
1321 done:
1322 	return rc;
1323 }
1324 
1325 static int decode_abs(struct x86_emulate_ctxt *ctxt,
1326 		      struct operand *op)
1327 {
1328 	int rc = X86EMUL_CONTINUE;
1329 
1330 	op->type = OP_MEM;
1331 	switch (ctxt->ad_bytes) {
1332 	case 2:
1333 		op->addr.mem.ea = insn_fetch(u16, ctxt);
1334 		break;
1335 	case 4:
1336 		op->addr.mem.ea = insn_fetch(u32, ctxt);
1337 		break;
1338 	case 8:
1339 		op->addr.mem.ea = insn_fetch(u64, ctxt);
1340 		break;
1341 	}
1342 done:
1343 	return rc;
1344 }
1345 
1346 static void fetch_bit_operand(struct x86_emulate_ctxt *ctxt)
1347 {
1348 	long sv = 0, mask;
1349 
1350 	if (ctxt->dst.type == OP_MEM && ctxt->src.type == OP_REG) {
1351 		mask = ~((long)ctxt->dst.bytes * 8 - 1);
1352 
1353 		if (ctxt->src.bytes == 2)
1354 			sv = (s16)ctxt->src.val & (s16)mask;
1355 		else if (ctxt->src.bytes == 4)
1356 			sv = (s32)ctxt->src.val & (s32)mask;
1357 		else
1358 			sv = (s64)ctxt->src.val & (s64)mask;
1359 
1360 		ctxt->dst.addr.mem.ea = address_mask(ctxt,
1361 					   ctxt->dst.addr.mem.ea + (sv >> 3));
1362 	}
1363 
1364 	/* only subword offset */
1365 	ctxt->src.val &= (ctxt->dst.bytes << 3) - 1;
1366 }
1367 
1368 static int read_emulated(struct x86_emulate_ctxt *ctxt,
1369 			 unsigned long addr, void *dest, unsigned size)
1370 {
1371 	int rc;
1372 	struct read_cache *mc = &ctxt->mem_read;
1373 
1374 	if (mc->pos < mc->end)
1375 		goto read_cached;
1376 
1377 	if (KVM_EMULATOR_BUG_ON((mc->end + size) >= sizeof(mc->data), ctxt))
1378 		return X86EMUL_UNHANDLEABLE;
1379 
1380 	rc = ctxt->ops->read_emulated(ctxt, addr, mc->data + mc->end, size,
1381 				      &ctxt->exception);
1382 	if (rc != X86EMUL_CONTINUE)
1383 		return rc;
1384 
1385 	mc->end += size;
1386 
1387 read_cached:
1388 	memcpy(dest, mc->data + mc->pos, size);
1389 	mc->pos += size;
1390 	return X86EMUL_CONTINUE;
1391 }
1392 
1393 static int segmented_read(struct x86_emulate_ctxt *ctxt,
1394 			  struct segmented_address addr,
1395 			  void *data,
1396 			  unsigned size)
1397 {
1398 	int rc;
1399 	ulong linear;
1400 
1401 	rc = linearize(ctxt, addr, size, false, &linear);
1402 	if (rc != X86EMUL_CONTINUE)
1403 		return rc;
1404 	return read_emulated(ctxt, linear, data, size);
1405 }
1406 
1407 static int segmented_write(struct x86_emulate_ctxt *ctxt,
1408 			   struct segmented_address addr,
1409 			   const void *data,
1410 			   unsigned size)
1411 {
1412 	int rc;
1413 	ulong linear;
1414 
1415 	rc = linearize(ctxt, addr, size, true, &linear);
1416 	if (rc != X86EMUL_CONTINUE)
1417 		return rc;
1418 	return ctxt->ops->write_emulated(ctxt, linear, data, size,
1419 					 &ctxt->exception);
1420 }
1421 
1422 static int segmented_cmpxchg(struct x86_emulate_ctxt *ctxt,
1423 			     struct segmented_address addr,
1424 			     const void *orig_data, const void *data,
1425 			     unsigned size)
1426 {
1427 	int rc;
1428 	ulong linear;
1429 
1430 	rc = linearize(ctxt, addr, size, true, &linear);
1431 	if (rc != X86EMUL_CONTINUE)
1432 		return rc;
1433 	return ctxt->ops->cmpxchg_emulated(ctxt, linear, orig_data, data,
1434 					   size, &ctxt->exception);
1435 }
1436 
1437 static int pio_in_emulated(struct x86_emulate_ctxt *ctxt,
1438 			   unsigned int size, unsigned short port,
1439 			   void *dest)
1440 {
1441 	struct read_cache *rc = &ctxt->io_read;
1442 
1443 	if (rc->pos == rc->end) { /* refill pio read ahead */
1444 		unsigned int in_page, n;
1445 		unsigned int count = ctxt->rep_prefix ?
1446 			address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) : 1;
1447 		in_page = (ctxt->eflags & X86_EFLAGS_DF) ?
1448 			offset_in_page(reg_read(ctxt, VCPU_REGS_RDI)) :
1449 			PAGE_SIZE - offset_in_page(reg_read(ctxt, VCPU_REGS_RDI));
1450 		n = min3(in_page, (unsigned int)sizeof(rc->data) / size, count);
1451 		if (n == 0)
1452 			n = 1;
1453 		rc->pos = rc->end = 0;
1454 		if (!ctxt->ops->pio_in_emulated(ctxt, size, port, rc->data, n))
1455 			return 0;
1456 		rc->end = n * size;
1457 	}
1458 
1459 	if (ctxt->rep_prefix && (ctxt->d & String) &&
1460 	    !(ctxt->eflags & X86_EFLAGS_DF)) {
1461 		ctxt->dst.data = rc->data + rc->pos;
1462 		ctxt->dst.type = OP_MEM_STR;
1463 		ctxt->dst.count = (rc->end - rc->pos) / size;
1464 		rc->pos = rc->end;
1465 	} else {
1466 		memcpy(dest, rc->data + rc->pos, size);
1467 		rc->pos += size;
1468 	}
1469 	return 1;
1470 }
1471 
1472 static int read_interrupt_descriptor(struct x86_emulate_ctxt *ctxt,
1473 				     u16 index, struct desc_struct *desc)
1474 {
1475 	struct desc_ptr dt;
1476 	ulong addr;
1477 
1478 	ctxt->ops->get_idt(ctxt, &dt);
1479 
1480 	if (dt.size < index * 8 + 7)
1481 		return emulate_gp(ctxt, index << 3 | 0x2);
1482 
1483 	addr = dt.address + index * 8;
1484 	return linear_read_system(ctxt, addr, desc, sizeof(*desc));
1485 }
1486 
1487 static void get_descriptor_table_ptr(struct x86_emulate_ctxt *ctxt,
1488 				     u16 selector, struct desc_ptr *dt)
1489 {
1490 	const struct x86_emulate_ops *ops = ctxt->ops;
1491 	u32 base3 = 0;
1492 
1493 	if (selector & 1 << 2) {
1494 		struct desc_struct desc;
1495 		u16 sel;
1496 
1497 		memset(dt, 0, sizeof(*dt));
1498 		if (!ops->get_segment(ctxt, &sel, &desc, &base3,
1499 				      VCPU_SREG_LDTR))
1500 			return;
1501 
1502 		dt->size = desc_limit_scaled(&desc); /* what if limit > 65535? */
1503 		dt->address = get_desc_base(&desc) | ((u64)base3 << 32);
1504 	} else
1505 		ops->get_gdt(ctxt, dt);
1506 }
1507 
1508 static int get_descriptor_ptr(struct x86_emulate_ctxt *ctxt,
1509 			      u16 selector, ulong *desc_addr_p)
1510 {
1511 	struct desc_ptr dt;
1512 	u16 index = selector >> 3;
1513 	ulong addr;
1514 
1515 	get_descriptor_table_ptr(ctxt, selector, &dt);
1516 
1517 	if (dt.size < index * 8 + 7)
1518 		return emulate_gp(ctxt, selector & 0xfffc);
1519 
1520 	addr = dt.address + index * 8;
1521 
1522 #ifdef CONFIG_X86_64
1523 	if (addr >> 32 != 0) {
1524 		u64 efer = 0;
1525 
1526 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1527 		if (!(efer & EFER_LMA))
1528 			addr &= (u32)-1;
1529 	}
1530 #endif
1531 
1532 	*desc_addr_p = addr;
1533 	return X86EMUL_CONTINUE;
1534 }
1535 
1536 /* allowed just for 8 bytes segments */
1537 static int read_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1538 				   u16 selector, struct desc_struct *desc,
1539 				   ulong *desc_addr_p)
1540 {
1541 	int rc;
1542 
1543 	rc = get_descriptor_ptr(ctxt, selector, desc_addr_p);
1544 	if (rc != X86EMUL_CONTINUE)
1545 		return rc;
1546 
1547 	return linear_read_system(ctxt, *desc_addr_p, desc, sizeof(*desc));
1548 }
1549 
1550 /* allowed just for 8 bytes segments */
1551 static int write_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1552 				    u16 selector, struct desc_struct *desc)
1553 {
1554 	int rc;
1555 	ulong addr;
1556 
1557 	rc = get_descriptor_ptr(ctxt, selector, &addr);
1558 	if (rc != X86EMUL_CONTINUE)
1559 		return rc;
1560 
1561 	return linear_write_system(ctxt, addr, desc, sizeof(*desc));
1562 }
1563 
1564 static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1565 				     u16 selector, int seg, u8 cpl,
1566 				     enum x86_transfer_type transfer,
1567 				     struct desc_struct *desc)
1568 {
1569 	struct desc_struct seg_desc, old_desc;
1570 	u8 dpl, rpl;
1571 	unsigned err_vec = GP_VECTOR;
1572 	u32 err_code = 0;
1573 	bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
1574 	ulong desc_addr;
1575 	int ret;
1576 	u16 dummy;
1577 	u32 base3 = 0;
1578 
1579 	memset(&seg_desc, 0, sizeof(seg_desc));
1580 
1581 	if (ctxt->mode == X86EMUL_MODE_REAL) {
1582 		/* set real mode segment descriptor (keep limit etc. for
1583 		 * unreal mode) */
1584 		ctxt->ops->get_segment(ctxt, &dummy, &seg_desc, NULL, seg);
1585 		set_desc_base(&seg_desc, selector << 4);
1586 		goto load;
1587 	} else if (seg <= VCPU_SREG_GS && ctxt->mode == X86EMUL_MODE_VM86) {
1588 		/* VM86 needs a clean new segment descriptor */
1589 		set_desc_base(&seg_desc, selector << 4);
1590 		set_desc_limit(&seg_desc, 0xffff);
1591 		seg_desc.type = 3;
1592 		seg_desc.p = 1;
1593 		seg_desc.s = 1;
1594 		seg_desc.dpl = 3;
1595 		goto load;
1596 	}
1597 
1598 	rpl = selector & 3;
1599 
1600 	/* TR should be in GDT only */
1601 	if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
1602 		goto exception;
1603 
1604 	/* NULL selector is not valid for TR, CS and (except for long mode) SS */
1605 	if (null_selector) {
1606 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_TR)
1607 			goto exception;
1608 
1609 		if (seg == VCPU_SREG_SS) {
1610 			if (ctxt->mode != X86EMUL_MODE_PROT64 || rpl != cpl)
1611 				goto exception;
1612 
1613 			/*
1614 			 * ctxt->ops->set_segment expects the CPL to be in
1615 			 * SS.DPL, so fake an expand-up 32-bit data segment.
1616 			 */
1617 			seg_desc.type = 3;
1618 			seg_desc.p = 1;
1619 			seg_desc.s = 1;
1620 			seg_desc.dpl = cpl;
1621 			seg_desc.d = 1;
1622 			seg_desc.g = 1;
1623 		}
1624 
1625 		/* Skip all following checks */
1626 		goto load;
1627 	}
1628 
1629 	ret = read_segment_descriptor(ctxt, selector, &seg_desc, &desc_addr);
1630 	if (ret != X86EMUL_CONTINUE)
1631 		return ret;
1632 
1633 	err_code = selector & 0xfffc;
1634 	err_vec = (transfer == X86_TRANSFER_TASK_SWITCH) ? TS_VECTOR :
1635 							   GP_VECTOR;
1636 
1637 	/* can't load system descriptor into segment selector */
1638 	if (seg <= VCPU_SREG_GS && !seg_desc.s) {
1639 		if (transfer == X86_TRANSFER_CALL_JMP)
1640 			return X86EMUL_UNHANDLEABLE;
1641 		goto exception;
1642 	}
1643 
1644 	dpl = seg_desc.dpl;
1645 
1646 	switch (seg) {
1647 	case VCPU_SREG_SS:
1648 		/*
1649 		 * segment is not a writable data segment or segment
1650 		 * selector's RPL != CPL or segment selector's RPL != CPL
1651 		 */
1652 		if (rpl != cpl || (seg_desc.type & 0xa) != 0x2 || dpl != cpl)
1653 			goto exception;
1654 		break;
1655 	case VCPU_SREG_CS:
1656 		if (!(seg_desc.type & 8))
1657 			goto exception;
1658 
1659 		if (transfer == X86_TRANSFER_RET) {
1660 			/* RET can never return to an inner privilege level. */
1661 			if (rpl < cpl)
1662 				goto exception;
1663 			/* Outer-privilege level return is not implemented */
1664 			if (rpl > cpl)
1665 				return X86EMUL_UNHANDLEABLE;
1666 		}
1667 		if (transfer == X86_TRANSFER_RET || transfer == X86_TRANSFER_TASK_SWITCH) {
1668 			if (seg_desc.type & 4) {
1669 				/* conforming */
1670 				if (dpl > rpl)
1671 					goto exception;
1672 			} else {
1673 				/* nonconforming */
1674 				if (dpl != rpl)
1675 					goto exception;
1676 			}
1677 		} else { /* X86_TRANSFER_CALL_JMP */
1678 			if (seg_desc.type & 4) {
1679 				/* conforming */
1680 				if (dpl > cpl)
1681 					goto exception;
1682 			} else {
1683 				/* nonconforming */
1684 				if (rpl > cpl || dpl != cpl)
1685 					goto exception;
1686 			}
1687 		}
1688 		/* in long-mode d/b must be clear if l is set */
1689 		if (seg_desc.d && seg_desc.l) {
1690 			u64 efer = 0;
1691 
1692 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1693 			if (efer & EFER_LMA)
1694 				goto exception;
1695 		}
1696 
1697 		/* CS(RPL) <- CPL */
1698 		selector = (selector & 0xfffc) | cpl;
1699 		break;
1700 	case VCPU_SREG_TR:
1701 		if (seg_desc.s || (seg_desc.type != 1 && seg_desc.type != 9))
1702 			goto exception;
1703 		break;
1704 	case VCPU_SREG_LDTR:
1705 		if (seg_desc.s || seg_desc.type != 2)
1706 			goto exception;
1707 		break;
1708 	default: /*  DS, ES, FS, or GS */
1709 		/*
1710 		 * segment is not a data or readable code segment or
1711 		 * ((segment is a data or nonconforming code segment)
1712 		 * and (both RPL and CPL > DPL))
1713 		 */
1714 		if ((seg_desc.type & 0xa) == 0x8 ||
1715 		    (((seg_desc.type & 0xc) != 0xc) &&
1716 		     (rpl > dpl && cpl > dpl)))
1717 			goto exception;
1718 		break;
1719 	}
1720 
1721 	if (!seg_desc.p) {
1722 		err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
1723 		goto exception;
1724 	}
1725 
1726 	if (seg_desc.s) {
1727 		/* mark segment as accessed */
1728 		if (!(seg_desc.type & 1)) {
1729 			seg_desc.type |= 1;
1730 			ret = write_segment_descriptor(ctxt, selector,
1731 						       &seg_desc);
1732 			if (ret != X86EMUL_CONTINUE)
1733 				return ret;
1734 		}
1735 	} else if (ctxt->mode == X86EMUL_MODE_PROT64) {
1736 		ret = linear_read_system(ctxt, desc_addr+8, &base3, sizeof(base3));
1737 		if (ret != X86EMUL_CONTINUE)
1738 			return ret;
1739 		if (emul_is_noncanonical_address(get_desc_base(&seg_desc) |
1740 						 ((u64)base3 << 32), ctxt))
1741 			return emulate_gp(ctxt, err_code);
1742 	}
1743 
1744 	if (seg == VCPU_SREG_TR) {
1745 		old_desc = seg_desc;
1746 		seg_desc.type |= 2; /* busy */
1747 		ret = ctxt->ops->cmpxchg_emulated(ctxt, desc_addr, &old_desc, &seg_desc,
1748 						  sizeof(seg_desc), &ctxt->exception);
1749 		if (ret != X86EMUL_CONTINUE)
1750 			return ret;
1751 	}
1752 load:
1753 	ctxt->ops->set_segment(ctxt, selector, &seg_desc, base3, seg);
1754 	if (desc)
1755 		*desc = seg_desc;
1756 	return X86EMUL_CONTINUE;
1757 exception:
1758 	return emulate_exception(ctxt, err_vec, err_code, true);
1759 }
1760 
1761 static int load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1762 				   u16 selector, int seg)
1763 {
1764 	u8 cpl = ctxt->ops->cpl(ctxt);
1765 
1766 	/*
1767 	 * None of MOV, POP and LSS can load a NULL selector in CPL=3, but
1768 	 * they can load it at CPL<3 (Intel's manual says only LSS can,
1769 	 * but it's wrong).
1770 	 *
1771 	 * However, the Intel manual says that putting IST=1/DPL=3 in
1772 	 * an interrupt gate will result in SS=3 (the AMD manual instead
1773 	 * says it doesn't), so allow SS=3 in __load_segment_descriptor
1774 	 * and only forbid it here.
1775 	 */
1776 	if (seg == VCPU_SREG_SS && selector == 3 &&
1777 	    ctxt->mode == X86EMUL_MODE_PROT64)
1778 		return emulate_exception(ctxt, GP_VECTOR, 0, true);
1779 
1780 	return __load_segment_descriptor(ctxt, selector, seg, cpl,
1781 					 X86_TRANSFER_NONE, NULL);
1782 }
1783 
1784 static void write_register_operand(struct operand *op)
1785 {
1786 	return assign_register(op->addr.reg, op->val, op->bytes);
1787 }
1788 
1789 static int writeback(struct x86_emulate_ctxt *ctxt, struct operand *op)
1790 {
1791 	switch (op->type) {
1792 	case OP_REG:
1793 		write_register_operand(op);
1794 		break;
1795 	case OP_MEM:
1796 		if (ctxt->lock_prefix)
1797 			return segmented_cmpxchg(ctxt,
1798 						 op->addr.mem,
1799 						 &op->orig_val,
1800 						 &op->val,
1801 						 op->bytes);
1802 		else
1803 			return segmented_write(ctxt,
1804 					       op->addr.mem,
1805 					       &op->val,
1806 					       op->bytes);
1807 		break;
1808 	case OP_MEM_STR:
1809 		return segmented_write(ctxt,
1810 				       op->addr.mem,
1811 				       op->data,
1812 				       op->bytes * op->count);
1813 		break;
1814 	case OP_XMM:
1815 		kvm_write_sse_reg(op->addr.xmm, &op->vec_val);
1816 		break;
1817 	case OP_MM:
1818 		kvm_write_mmx_reg(op->addr.mm, &op->mm_val);
1819 		break;
1820 	case OP_NONE:
1821 		/* no writeback */
1822 		break;
1823 	default:
1824 		break;
1825 	}
1826 	return X86EMUL_CONTINUE;
1827 }
1828 
1829 static int push(struct x86_emulate_ctxt *ctxt, void *data, int bytes)
1830 {
1831 	struct segmented_address addr;
1832 
1833 	rsp_increment(ctxt, -bytes);
1834 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1835 	addr.seg = VCPU_SREG_SS;
1836 
1837 	return segmented_write(ctxt, addr, data, bytes);
1838 }
1839 
1840 static int em_push(struct x86_emulate_ctxt *ctxt)
1841 {
1842 	/* Disable writeback. */
1843 	ctxt->dst.type = OP_NONE;
1844 	return push(ctxt, &ctxt->src.val, ctxt->op_bytes);
1845 }
1846 
1847 static int emulate_pop(struct x86_emulate_ctxt *ctxt,
1848 		       void *dest, int len)
1849 {
1850 	int rc;
1851 	struct segmented_address addr;
1852 
1853 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1854 	addr.seg = VCPU_SREG_SS;
1855 	rc = segmented_read(ctxt, addr, dest, len);
1856 	if (rc != X86EMUL_CONTINUE)
1857 		return rc;
1858 
1859 	rsp_increment(ctxt, len);
1860 	return rc;
1861 }
1862 
1863 static int em_pop(struct x86_emulate_ctxt *ctxt)
1864 {
1865 	return emulate_pop(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1866 }
1867 
1868 static int emulate_popf(struct x86_emulate_ctxt *ctxt,
1869 			void *dest, int len)
1870 {
1871 	int rc;
1872 	unsigned long val, change_mask;
1873 	int iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
1874 	int cpl = ctxt->ops->cpl(ctxt);
1875 
1876 	rc = emulate_pop(ctxt, &val, len);
1877 	if (rc != X86EMUL_CONTINUE)
1878 		return rc;
1879 
1880 	change_mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
1881 		      X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF |
1882 		      X86_EFLAGS_TF | X86_EFLAGS_DF | X86_EFLAGS_NT |
1883 		      X86_EFLAGS_AC | X86_EFLAGS_ID;
1884 
1885 	switch(ctxt->mode) {
1886 	case X86EMUL_MODE_PROT64:
1887 	case X86EMUL_MODE_PROT32:
1888 	case X86EMUL_MODE_PROT16:
1889 		if (cpl == 0)
1890 			change_mask |= X86_EFLAGS_IOPL;
1891 		if (cpl <= iopl)
1892 			change_mask |= X86_EFLAGS_IF;
1893 		break;
1894 	case X86EMUL_MODE_VM86:
1895 		if (iopl < 3)
1896 			return emulate_gp(ctxt, 0);
1897 		change_mask |= X86_EFLAGS_IF;
1898 		break;
1899 	default: /* real mode */
1900 		change_mask |= (X86_EFLAGS_IOPL | X86_EFLAGS_IF);
1901 		break;
1902 	}
1903 
1904 	*(unsigned long *)dest =
1905 		(ctxt->eflags & ~change_mask) | (val & change_mask);
1906 
1907 	return rc;
1908 }
1909 
1910 static int em_popf(struct x86_emulate_ctxt *ctxt)
1911 {
1912 	ctxt->dst.type = OP_REG;
1913 	ctxt->dst.addr.reg = &ctxt->eflags;
1914 	ctxt->dst.bytes = ctxt->op_bytes;
1915 	return emulate_popf(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1916 }
1917 
1918 static int em_enter(struct x86_emulate_ctxt *ctxt)
1919 {
1920 	int rc;
1921 	unsigned frame_size = ctxt->src.val;
1922 	unsigned nesting_level = ctxt->src2.val & 31;
1923 	ulong rbp;
1924 
1925 	if (nesting_level)
1926 		return X86EMUL_UNHANDLEABLE;
1927 
1928 	rbp = reg_read(ctxt, VCPU_REGS_RBP);
1929 	rc = push(ctxt, &rbp, stack_size(ctxt));
1930 	if (rc != X86EMUL_CONTINUE)
1931 		return rc;
1932 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RBP), reg_read(ctxt, VCPU_REGS_RSP),
1933 		      stack_mask(ctxt));
1934 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP),
1935 		      reg_read(ctxt, VCPU_REGS_RSP) - frame_size,
1936 		      stack_mask(ctxt));
1937 	return X86EMUL_CONTINUE;
1938 }
1939 
1940 static int em_leave(struct x86_emulate_ctxt *ctxt)
1941 {
1942 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP), reg_read(ctxt, VCPU_REGS_RBP),
1943 		      stack_mask(ctxt));
1944 	return emulate_pop(ctxt, reg_rmw(ctxt, VCPU_REGS_RBP), ctxt->op_bytes);
1945 }
1946 
1947 static int em_push_sreg(struct x86_emulate_ctxt *ctxt)
1948 {
1949 	int seg = ctxt->src2.val;
1950 
1951 	ctxt->src.val = get_segment_selector(ctxt, seg);
1952 	if (ctxt->op_bytes == 4) {
1953 		rsp_increment(ctxt, -2);
1954 		ctxt->op_bytes = 2;
1955 	}
1956 
1957 	return em_push(ctxt);
1958 }
1959 
1960 static int em_pop_sreg(struct x86_emulate_ctxt *ctxt)
1961 {
1962 	int seg = ctxt->src2.val;
1963 	unsigned long selector;
1964 	int rc;
1965 
1966 	rc = emulate_pop(ctxt, &selector, 2);
1967 	if (rc != X86EMUL_CONTINUE)
1968 		return rc;
1969 
1970 	if (ctxt->modrm_reg == VCPU_SREG_SS)
1971 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
1972 	if (ctxt->op_bytes > 2)
1973 		rsp_increment(ctxt, ctxt->op_bytes - 2);
1974 
1975 	rc = load_segment_descriptor(ctxt, (u16)selector, seg);
1976 	return rc;
1977 }
1978 
1979 static int em_pusha(struct x86_emulate_ctxt *ctxt)
1980 {
1981 	unsigned long old_esp = reg_read(ctxt, VCPU_REGS_RSP);
1982 	int rc = X86EMUL_CONTINUE;
1983 	int reg = VCPU_REGS_RAX;
1984 
1985 	while (reg <= VCPU_REGS_RDI) {
1986 		(reg == VCPU_REGS_RSP) ?
1987 		(ctxt->src.val = old_esp) : (ctxt->src.val = reg_read(ctxt, reg));
1988 
1989 		rc = em_push(ctxt);
1990 		if (rc != X86EMUL_CONTINUE)
1991 			return rc;
1992 
1993 		++reg;
1994 	}
1995 
1996 	return rc;
1997 }
1998 
1999 static int em_pushf(struct x86_emulate_ctxt *ctxt)
2000 {
2001 	ctxt->src.val = (unsigned long)ctxt->eflags & ~X86_EFLAGS_VM;
2002 	return em_push(ctxt);
2003 }
2004 
2005 static int em_popa(struct x86_emulate_ctxt *ctxt)
2006 {
2007 	int rc = X86EMUL_CONTINUE;
2008 	int reg = VCPU_REGS_RDI;
2009 	u32 val;
2010 
2011 	while (reg >= VCPU_REGS_RAX) {
2012 		if (reg == VCPU_REGS_RSP) {
2013 			rsp_increment(ctxt, ctxt->op_bytes);
2014 			--reg;
2015 		}
2016 
2017 		rc = emulate_pop(ctxt, &val, ctxt->op_bytes);
2018 		if (rc != X86EMUL_CONTINUE)
2019 			break;
2020 		assign_register(reg_rmw(ctxt, reg), val, ctxt->op_bytes);
2021 		--reg;
2022 	}
2023 	return rc;
2024 }
2025 
2026 static int __emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2027 {
2028 	const struct x86_emulate_ops *ops = ctxt->ops;
2029 	int rc;
2030 	struct desc_ptr dt;
2031 	gva_t cs_addr;
2032 	gva_t eip_addr;
2033 	u16 cs, eip;
2034 
2035 	/* TODO: Add limit checks */
2036 	ctxt->src.val = ctxt->eflags;
2037 	rc = em_push(ctxt);
2038 	if (rc != X86EMUL_CONTINUE)
2039 		return rc;
2040 
2041 	ctxt->eflags &= ~(X86_EFLAGS_IF | X86_EFLAGS_TF | X86_EFLAGS_AC);
2042 
2043 	ctxt->src.val = get_segment_selector(ctxt, VCPU_SREG_CS);
2044 	rc = em_push(ctxt);
2045 	if (rc != X86EMUL_CONTINUE)
2046 		return rc;
2047 
2048 	ctxt->src.val = ctxt->_eip;
2049 	rc = em_push(ctxt);
2050 	if (rc != X86EMUL_CONTINUE)
2051 		return rc;
2052 
2053 	ops->get_idt(ctxt, &dt);
2054 
2055 	eip_addr = dt.address + (irq << 2);
2056 	cs_addr = dt.address + (irq << 2) + 2;
2057 
2058 	rc = linear_read_system(ctxt, cs_addr, &cs, 2);
2059 	if (rc != X86EMUL_CONTINUE)
2060 		return rc;
2061 
2062 	rc = linear_read_system(ctxt, eip_addr, &eip, 2);
2063 	if (rc != X86EMUL_CONTINUE)
2064 		return rc;
2065 
2066 	rc = load_segment_descriptor(ctxt, cs, VCPU_SREG_CS);
2067 	if (rc != X86EMUL_CONTINUE)
2068 		return rc;
2069 
2070 	ctxt->_eip = eip;
2071 
2072 	return rc;
2073 }
2074 
2075 int emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2076 {
2077 	int rc;
2078 
2079 	invalidate_registers(ctxt);
2080 	rc = __emulate_int_real(ctxt, irq);
2081 	if (rc == X86EMUL_CONTINUE)
2082 		writeback_registers(ctxt);
2083 	return rc;
2084 }
2085 
2086 static int emulate_int(struct x86_emulate_ctxt *ctxt, int irq)
2087 {
2088 	switch(ctxt->mode) {
2089 	case X86EMUL_MODE_REAL:
2090 		return __emulate_int_real(ctxt, irq);
2091 	case X86EMUL_MODE_VM86:
2092 	case X86EMUL_MODE_PROT16:
2093 	case X86EMUL_MODE_PROT32:
2094 	case X86EMUL_MODE_PROT64:
2095 	default:
2096 		/* Protected mode interrupts unimplemented yet */
2097 		return X86EMUL_UNHANDLEABLE;
2098 	}
2099 }
2100 
2101 static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
2102 {
2103 	int rc = X86EMUL_CONTINUE;
2104 	unsigned long temp_eip = 0;
2105 	unsigned long temp_eflags = 0;
2106 	unsigned long cs = 0;
2107 	unsigned long mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
2108 			     X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_TF |
2109 			     X86_EFLAGS_IF | X86_EFLAGS_DF | X86_EFLAGS_OF |
2110 			     X86_EFLAGS_IOPL | X86_EFLAGS_NT | X86_EFLAGS_RF |
2111 			     X86_EFLAGS_AC | X86_EFLAGS_ID |
2112 			     X86_EFLAGS_FIXED;
2113 	unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF |
2114 				  X86_EFLAGS_VIP;
2115 
2116 	/* TODO: Add stack limit check */
2117 
2118 	rc = emulate_pop(ctxt, &temp_eip, ctxt->op_bytes);
2119 
2120 	if (rc != X86EMUL_CONTINUE)
2121 		return rc;
2122 
2123 	if (temp_eip & ~0xffff)
2124 		return emulate_gp(ctxt, 0);
2125 
2126 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2127 
2128 	if (rc != X86EMUL_CONTINUE)
2129 		return rc;
2130 
2131 	rc = emulate_pop(ctxt, &temp_eflags, ctxt->op_bytes);
2132 
2133 	if (rc != X86EMUL_CONTINUE)
2134 		return rc;
2135 
2136 	rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS);
2137 
2138 	if (rc != X86EMUL_CONTINUE)
2139 		return rc;
2140 
2141 	ctxt->_eip = temp_eip;
2142 
2143 	if (ctxt->op_bytes == 4)
2144 		ctxt->eflags = ((temp_eflags & mask) | (ctxt->eflags & vm86_mask));
2145 	else if (ctxt->op_bytes == 2) {
2146 		ctxt->eflags &= ~0xffff;
2147 		ctxt->eflags |= temp_eflags;
2148 	}
2149 
2150 	ctxt->eflags &= ~EFLG_RESERVED_ZEROS_MASK; /* Clear reserved zeros */
2151 	ctxt->eflags |= X86_EFLAGS_FIXED;
2152 	ctxt->ops->set_nmi_mask(ctxt, false);
2153 
2154 	return rc;
2155 }
2156 
2157 static int em_iret(struct x86_emulate_ctxt *ctxt)
2158 {
2159 	switch(ctxt->mode) {
2160 	case X86EMUL_MODE_REAL:
2161 		return emulate_iret_real(ctxt);
2162 	case X86EMUL_MODE_VM86:
2163 	case X86EMUL_MODE_PROT16:
2164 	case X86EMUL_MODE_PROT32:
2165 	case X86EMUL_MODE_PROT64:
2166 	default:
2167 		/* iret from protected mode unimplemented yet */
2168 		return X86EMUL_UNHANDLEABLE;
2169 	}
2170 }
2171 
2172 static int em_jmp_far(struct x86_emulate_ctxt *ctxt)
2173 {
2174 	int rc;
2175 	unsigned short sel;
2176 	struct desc_struct new_desc;
2177 	u8 cpl = ctxt->ops->cpl(ctxt);
2178 
2179 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2180 
2181 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
2182 				       X86_TRANSFER_CALL_JMP,
2183 				       &new_desc);
2184 	if (rc != X86EMUL_CONTINUE)
2185 		return rc;
2186 
2187 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
2188 	/* Error handling is not implemented. */
2189 	if (rc != X86EMUL_CONTINUE)
2190 		return X86EMUL_UNHANDLEABLE;
2191 
2192 	return rc;
2193 }
2194 
2195 static int em_jmp_abs(struct x86_emulate_ctxt *ctxt)
2196 {
2197 	return assign_eip_near(ctxt, ctxt->src.val);
2198 }
2199 
2200 static int em_call_near_abs(struct x86_emulate_ctxt *ctxt)
2201 {
2202 	int rc;
2203 	long int old_eip;
2204 
2205 	old_eip = ctxt->_eip;
2206 	rc = assign_eip_near(ctxt, ctxt->src.val);
2207 	if (rc != X86EMUL_CONTINUE)
2208 		return rc;
2209 	ctxt->src.val = old_eip;
2210 	rc = em_push(ctxt);
2211 	return rc;
2212 }
2213 
2214 static int em_cmpxchg8b(struct x86_emulate_ctxt *ctxt)
2215 {
2216 	u64 old = ctxt->dst.orig_val64;
2217 
2218 	if (ctxt->dst.bytes == 16)
2219 		return X86EMUL_UNHANDLEABLE;
2220 
2221 	if (((u32) (old >> 0) != (u32) reg_read(ctxt, VCPU_REGS_RAX)) ||
2222 	    ((u32) (old >> 32) != (u32) reg_read(ctxt, VCPU_REGS_RDX))) {
2223 		*reg_write(ctxt, VCPU_REGS_RAX) = (u32) (old >> 0);
2224 		*reg_write(ctxt, VCPU_REGS_RDX) = (u32) (old >> 32);
2225 		ctxt->eflags &= ~X86_EFLAGS_ZF;
2226 	} else {
2227 		ctxt->dst.val64 = ((u64)reg_read(ctxt, VCPU_REGS_RCX) << 32) |
2228 			(u32) reg_read(ctxt, VCPU_REGS_RBX);
2229 
2230 		ctxt->eflags |= X86_EFLAGS_ZF;
2231 	}
2232 	return X86EMUL_CONTINUE;
2233 }
2234 
2235 static int em_ret(struct x86_emulate_ctxt *ctxt)
2236 {
2237 	int rc;
2238 	unsigned long eip;
2239 
2240 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2241 	if (rc != X86EMUL_CONTINUE)
2242 		return rc;
2243 
2244 	return assign_eip_near(ctxt, eip);
2245 }
2246 
2247 static int em_ret_far(struct x86_emulate_ctxt *ctxt)
2248 {
2249 	int rc;
2250 	unsigned long eip, cs;
2251 	int cpl = ctxt->ops->cpl(ctxt);
2252 	struct desc_struct new_desc;
2253 
2254 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2255 	if (rc != X86EMUL_CONTINUE)
2256 		return rc;
2257 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2258 	if (rc != X86EMUL_CONTINUE)
2259 		return rc;
2260 	rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl,
2261 				       X86_TRANSFER_RET,
2262 				       &new_desc);
2263 	if (rc != X86EMUL_CONTINUE)
2264 		return rc;
2265 	rc = assign_eip_far(ctxt, eip, &new_desc);
2266 	/* Error handling is not implemented. */
2267 	if (rc != X86EMUL_CONTINUE)
2268 		return X86EMUL_UNHANDLEABLE;
2269 
2270 	return rc;
2271 }
2272 
2273 static int em_ret_far_imm(struct x86_emulate_ctxt *ctxt)
2274 {
2275         int rc;
2276 
2277         rc = em_ret_far(ctxt);
2278         if (rc != X86EMUL_CONTINUE)
2279                 return rc;
2280         rsp_increment(ctxt, ctxt->src.val);
2281         return X86EMUL_CONTINUE;
2282 }
2283 
2284 static int em_cmpxchg(struct x86_emulate_ctxt *ctxt)
2285 {
2286 	/* Save real source value, then compare EAX against destination. */
2287 	ctxt->dst.orig_val = ctxt->dst.val;
2288 	ctxt->dst.val = reg_read(ctxt, VCPU_REGS_RAX);
2289 	ctxt->src.orig_val = ctxt->src.val;
2290 	ctxt->src.val = ctxt->dst.orig_val;
2291 	fastop(ctxt, em_cmp);
2292 
2293 	if (ctxt->eflags & X86_EFLAGS_ZF) {
2294 		/* Success: write back to memory; no update of EAX */
2295 		ctxt->src.type = OP_NONE;
2296 		ctxt->dst.val = ctxt->src.orig_val;
2297 	} else {
2298 		/* Failure: write the value we saw to EAX. */
2299 		ctxt->src.type = OP_REG;
2300 		ctxt->src.addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
2301 		ctxt->src.val = ctxt->dst.orig_val;
2302 		/* Create write-cycle to dest by writing the same value */
2303 		ctxt->dst.val = ctxt->dst.orig_val;
2304 	}
2305 	return X86EMUL_CONTINUE;
2306 }
2307 
2308 static int em_lseg(struct x86_emulate_ctxt *ctxt)
2309 {
2310 	int seg = ctxt->src2.val;
2311 	unsigned short sel;
2312 	int rc;
2313 
2314 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2315 
2316 	rc = load_segment_descriptor(ctxt, sel, seg);
2317 	if (rc != X86EMUL_CONTINUE)
2318 		return rc;
2319 
2320 	ctxt->dst.val = ctxt->src.val;
2321 	return rc;
2322 }
2323 
2324 static int emulator_has_longmode(struct x86_emulate_ctxt *ctxt)
2325 {
2326 #ifdef CONFIG_X86_64
2327 	return ctxt->ops->guest_has_long_mode(ctxt);
2328 #else
2329 	return false;
2330 #endif
2331 }
2332 
2333 static void rsm_set_desc_flags(struct desc_struct *desc, u32 flags)
2334 {
2335 	desc->g    = (flags >> 23) & 1;
2336 	desc->d    = (flags >> 22) & 1;
2337 	desc->l    = (flags >> 21) & 1;
2338 	desc->avl  = (flags >> 20) & 1;
2339 	desc->p    = (flags >> 15) & 1;
2340 	desc->dpl  = (flags >> 13) & 3;
2341 	desc->s    = (flags >> 12) & 1;
2342 	desc->type = (flags >>  8) & 15;
2343 }
2344 
2345 static int rsm_load_seg_32(struct x86_emulate_ctxt *ctxt, const char *smstate,
2346 			   int n)
2347 {
2348 	struct desc_struct desc;
2349 	int offset;
2350 	u16 selector;
2351 
2352 	selector = GET_SMSTATE(u32, smstate, 0x7fa8 + n * 4);
2353 
2354 	if (n < 3)
2355 		offset = 0x7f84 + n * 12;
2356 	else
2357 		offset = 0x7f2c + (n - 3) * 12;
2358 
2359 	set_desc_base(&desc,      GET_SMSTATE(u32, smstate, offset + 8));
2360 	set_desc_limit(&desc,     GET_SMSTATE(u32, smstate, offset + 4));
2361 	rsm_set_desc_flags(&desc, GET_SMSTATE(u32, smstate, offset));
2362 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, n);
2363 	return X86EMUL_CONTINUE;
2364 }
2365 
2366 #ifdef CONFIG_X86_64
2367 static int rsm_load_seg_64(struct x86_emulate_ctxt *ctxt, const char *smstate,
2368 			   int n)
2369 {
2370 	struct desc_struct desc;
2371 	int offset;
2372 	u16 selector;
2373 	u32 base3;
2374 
2375 	offset = 0x7e00 + n * 16;
2376 
2377 	selector =                GET_SMSTATE(u16, smstate, offset);
2378 	rsm_set_desc_flags(&desc, GET_SMSTATE(u16, smstate, offset + 2) << 8);
2379 	set_desc_limit(&desc,     GET_SMSTATE(u32, smstate, offset + 4));
2380 	set_desc_base(&desc,      GET_SMSTATE(u32, smstate, offset + 8));
2381 	base3 =                   GET_SMSTATE(u32, smstate, offset + 12);
2382 
2383 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, n);
2384 	return X86EMUL_CONTINUE;
2385 }
2386 #endif
2387 
2388 static int rsm_enter_protected_mode(struct x86_emulate_ctxt *ctxt,
2389 				    u64 cr0, u64 cr3, u64 cr4)
2390 {
2391 	int bad;
2392 	u64 pcid;
2393 
2394 	/* In order to later set CR4.PCIDE, CR3[11:0] must be zero.  */
2395 	pcid = 0;
2396 	if (cr4 & X86_CR4_PCIDE) {
2397 		pcid = cr3 & 0xfff;
2398 		cr3 &= ~0xfff;
2399 	}
2400 
2401 	bad = ctxt->ops->set_cr(ctxt, 3, cr3);
2402 	if (bad)
2403 		return X86EMUL_UNHANDLEABLE;
2404 
2405 	/*
2406 	 * First enable PAE, long mode needs it before CR0.PG = 1 is set.
2407 	 * Then enable protected mode.	However, PCID cannot be enabled
2408 	 * if EFER.LMA=0, so set it separately.
2409 	 */
2410 	bad = ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2411 	if (bad)
2412 		return X86EMUL_UNHANDLEABLE;
2413 
2414 	bad = ctxt->ops->set_cr(ctxt, 0, cr0);
2415 	if (bad)
2416 		return X86EMUL_UNHANDLEABLE;
2417 
2418 	if (cr4 & X86_CR4_PCIDE) {
2419 		bad = ctxt->ops->set_cr(ctxt, 4, cr4);
2420 		if (bad)
2421 			return X86EMUL_UNHANDLEABLE;
2422 		if (pcid) {
2423 			bad = ctxt->ops->set_cr(ctxt, 3, cr3 | pcid);
2424 			if (bad)
2425 				return X86EMUL_UNHANDLEABLE;
2426 		}
2427 
2428 	}
2429 
2430 	return X86EMUL_CONTINUE;
2431 }
2432 
2433 static int rsm_load_state_32(struct x86_emulate_ctxt *ctxt,
2434 			     const char *smstate)
2435 {
2436 	struct desc_struct desc;
2437 	struct desc_ptr dt;
2438 	u16 selector;
2439 	u32 val, cr0, cr3, cr4;
2440 	int i;
2441 
2442 	cr0 =                      GET_SMSTATE(u32, smstate, 0x7ffc);
2443 	cr3 =                      GET_SMSTATE(u32, smstate, 0x7ff8);
2444 	ctxt->eflags =             GET_SMSTATE(u32, smstate, 0x7ff4) | X86_EFLAGS_FIXED;
2445 	ctxt->_eip =               GET_SMSTATE(u32, smstate, 0x7ff0);
2446 
2447 	for (i = 0; i < NR_EMULATOR_GPRS; i++)
2448 		*reg_write(ctxt, i) = GET_SMSTATE(u32, smstate, 0x7fd0 + i * 4);
2449 
2450 	val = GET_SMSTATE(u32, smstate, 0x7fcc);
2451 
2452 	if (ctxt->ops->set_dr(ctxt, 6, val))
2453 		return X86EMUL_UNHANDLEABLE;
2454 
2455 	val = GET_SMSTATE(u32, smstate, 0x7fc8);
2456 
2457 	if (ctxt->ops->set_dr(ctxt, 7, val))
2458 		return X86EMUL_UNHANDLEABLE;
2459 
2460 	selector =                 GET_SMSTATE(u32, smstate, 0x7fc4);
2461 	set_desc_base(&desc,       GET_SMSTATE(u32, smstate, 0x7f64));
2462 	set_desc_limit(&desc,      GET_SMSTATE(u32, smstate, 0x7f60));
2463 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smstate, 0x7f5c));
2464 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_TR);
2465 
2466 	selector =                 GET_SMSTATE(u32, smstate, 0x7fc0);
2467 	set_desc_base(&desc,       GET_SMSTATE(u32, smstate, 0x7f80));
2468 	set_desc_limit(&desc,      GET_SMSTATE(u32, smstate, 0x7f7c));
2469 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smstate, 0x7f78));
2470 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_LDTR);
2471 
2472 	dt.address =               GET_SMSTATE(u32, smstate, 0x7f74);
2473 	dt.size =                  GET_SMSTATE(u32, smstate, 0x7f70);
2474 	ctxt->ops->set_gdt(ctxt, &dt);
2475 
2476 	dt.address =               GET_SMSTATE(u32, smstate, 0x7f58);
2477 	dt.size =                  GET_SMSTATE(u32, smstate, 0x7f54);
2478 	ctxt->ops->set_idt(ctxt, &dt);
2479 
2480 	for (i = 0; i < 6; i++) {
2481 		int r = rsm_load_seg_32(ctxt, smstate, i);
2482 		if (r != X86EMUL_CONTINUE)
2483 			return r;
2484 	}
2485 
2486 	cr4 = GET_SMSTATE(u32, smstate, 0x7f14);
2487 
2488 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smstate, 0x7ef8));
2489 
2490 	return rsm_enter_protected_mode(ctxt, cr0, cr3, cr4);
2491 }
2492 
2493 #ifdef CONFIG_X86_64
2494 static int rsm_load_state_64(struct x86_emulate_ctxt *ctxt,
2495 			     const char *smstate)
2496 {
2497 	struct desc_struct desc;
2498 	struct desc_ptr dt;
2499 	u64 val, cr0, cr3, cr4;
2500 	u32 base3;
2501 	u16 selector;
2502 	int i, r;
2503 
2504 	for (i = 0; i < NR_EMULATOR_GPRS; i++)
2505 		*reg_write(ctxt, i) = GET_SMSTATE(u64, smstate, 0x7ff8 - i * 8);
2506 
2507 	ctxt->_eip   = GET_SMSTATE(u64, smstate, 0x7f78);
2508 	ctxt->eflags = GET_SMSTATE(u32, smstate, 0x7f70) | X86_EFLAGS_FIXED;
2509 
2510 	val = GET_SMSTATE(u64, smstate, 0x7f68);
2511 
2512 	if (ctxt->ops->set_dr(ctxt, 6, val))
2513 		return X86EMUL_UNHANDLEABLE;
2514 
2515 	val = GET_SMSTATE(u64, smstate, 0x7f60);
2516 
2517 	if (ctxt->ops->set_dr(ctxt, 7, val))
2518 		return X86EMUL_UNHANDLEABLE;
2519 
2520 	cr0 =                       GET_SMSTATE(u64, smstate, 0x7f58);
2521 	cr3 =                       GET_SMSTATE(u64, smstate, 0x7f50);
2522 	cr4 =                       GET_SMSTATE(u64, smstate, 0x7f48);
2523 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smstate, 0x7f00));
2524 	val =                       GET_SMSTATE(u64, smstate, 0x7ed0);
2525 
2526 	if (ctxt->ops->set_msr(ctxt, MSR_EFER, val & ~EFER_LMA))
2527 		return X86EMUL_UNHANDLEABLE;
2528 
2529 	selector =                  GET_SMSTATE(u32, smstate, 0x7e90);
2530 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smstate, 0x7e92) << 8);
2531 	set_desc_limit(&desc,       GET_SMSTATE(u32, smstate, 0x7e94));
2532 	set_desc_base(&desc,        GET_SMSTATE(u32, smstate, 0x7e98));
2533 	base3 =                     GET_SMSTATE(u32, smstate, 0x7e9c);
2534 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_TR);
2535 
2536 	dt.size =                   GET_SMSTATE(u32, smstate, 0x7e84);
2537 	dt.address =                GET_SMSTATE(u64, smstate, 0x7e88);
2538 	ctxt->ops->set_idt(ctxt, &dt);
2539 
2540 	selector =                  GET_SMSTATE(u32, smstate, 0x7e70);
2541 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smstate, 0x7e72) << 8);
2542 	set_desc_limit(&desc,       GET_SMSTATE(u32, smstate, 0x7e74));
2543 	set_desc_base(&desc,        GET_SMSTATE(u32, smstate, 0x7e78));
2544 	base3 =                     GET_SMSTATE(u32, smstate, 0x7e7c);
2545 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_LDTR);
2546 
2547 	dt.size =                   GET_SMSTATE(u32, smstate, 0x7e64);
2548 	dt.address =                GET_SMSTATE(u64, smstate, 0x7e68);
2549 	ctxt->ops->set_gdt(ctxt, &dt);
2550 
2551 	r = rsm_enter_protected_mode(ctxt, cr0, cr3, cr4);
2552 	if (r != X86EMUL_CONTINUE)
2553 		return r;
2554 
2555 	for (i = 0; i < 6; i++) {
2556 		r = rsm_load_seg_64(ctxt, smstate, i);
2557 		if (r != X86EMUL_CONTINUE)
2558 			return r;
2559 	}
2560 
2561 	return X86EMUL_CONTINUE;
2562 }
2563 #endif
2564 
2565 static int em_rsm(struct x86_emulate_ctxt *ctxt)
2566 {
2567 	unsigned long cr0, cr4, efer;
2568 	char buf[512];
2569 	u64 smbase;
2570 	int ret;
2571 
2572 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_MASK) == 0)
2573 		return emulate_ud(ctxt);
2574 
2575 	smbase = ctxt->ops->get_smbase(ctxt);
2576 
2577 	ret = ctxt->ops->read_phys(ctxt, smbase + 0xfe00, buf, sizeof(buf));
2578 	if (ret != X86EMUL_CONTINUE)
2579 		return X86EMUL_UNHANDLEABLE;
2580 
2581 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_INSIDE_NMI_MASK) == 0)
2582 		ctxt->ops->set_nmi_mask(ctxt, false);
2583 
2584 	ctxt->ops->exiting_smm(ctxt);
2585 
2586 	/*
2587 	 * Get back to real mode, to prepare a safe state in which to load
2588 	 * CR0/CR3/CR4/EFER.  It's all a bit more complicated if the vCPU
2589 	 * supports long mode.
2590 	 */
2591 	if (emulator_has_longmode(ctxt)) {
2592 		struct desc_struct cs_desc;
2593 
2594 		/* Zero CR4.PCIDE before CR0.PG.  */
2595 		cr4 = ctxt->ops->get_cr(ctxt, 4);
2596 		if (cr4 & X86_CR4_PCIDE)
2597 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2598 
2599 		/* A 32-bit code segment is required to clear EFER.LMA.  */
2600 		memset(&cs_desc, 0, sizeof(cs_desc));
2601 		cs_desc.type = 0xb;
2602 		cs_desc.s = cs_desc.g = cs_desc.p = 1;
2603 		ctxt->ops->set_segment(ctxt, 0, &cs_desc, 0, VCPU_SREG_CS);
2604 	}
2605 
2606 	/* For the 64-bit case, this will clear EFER.LMA.  */
2607 	cr0 = ctxt->ops->get_cr(ctxt, 0);
2608 	if (cr0 & X86_CR0_PE)
2609 		ctxt->ops->set_cr(ctxt, 0, cr0 & ~(X86_CR0_PG | X86_CR0_PE));
2610 
2611 	if (emulator_has_longmode(ctxt)) {
2612 		/* Clear CR4.PAE before clearing EFER.LME. */
2613 		cr4 = ctxt->ops->get_cr(ctxt, 4);
2614 		if (cr4 & X86_CR4_PAE)
2615 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PAE);
2616 
2617 		/* And finally go back to 32-bit mode.  */
2618 		efer = 0;
2619 		ctxt->ops->set_msr(ctxt, MSR_EFER, efer);
2620 	}
2621 
2622 	/*
2623 	 * Give leave_smm() a chance to make ISA-specific changes to the vCPU
2624 	 * state (e.g. enter guest mode) before loading state from the SMM
2625 	 * state-save area.
2626 	 */
2627 	if (ctxt->ops->leave_smm(ctxt, buf))
2628 		goto emulate_shutdown;
2629 
2630 #ifdef CONFIG_X86_64
2631 	if (emulator_has_longmode(ctxt))
2632 		ret = rsm_load_state_64(ctxt, buf);
2633 	else
2634 #endif
2635 		ret = rsm_load_state_32(ctxt, buf);
2636 
2637 	if (ret != X86EMUL_CONTINUE)
2638 		goto emulate_shutdown;
2639 
2640 	/*
2641 	 * Note, the ctxt->ops callbacks are responsible for handling side
2642 	 * effects when writing MSRs and CRs, e.g. MMU context resets, CPUID
2643 	 * runtime updates, etc...  If that changes, e.g. this flow is moved
2644 	 * out of the emulator to make it look more like enter_smm(), then
2645 	 * those side effects need to be explicitly handled for both success
2646 	 * and shutdown.
2647 	 */
2648 	return X86EMUL_CONTINUE;
2649 
2650 emulate_shutdown:
2651 	ctxt->ops->triple_fault(ctxt);
2652 	return X86EMUL_CONTINUE;
2653 }
2654 
2655 static void
2656 setup_syscalls_segments(struct desc_struct *cs, struct desc_struct *ss)
2657 {
2658 	cs->l = 0;		/* will be adjusted later */
2659 	set_desc_base(cs, 0);	/* flat segment */
2660 	cs->g = 1;		/* 4kb granularity */
2661 	set_desc_limit(cs, 0xfffff);	/* 4GB limit */
2662 	cs->type = 0x0b;	/* Read, Execute, Accessed */
2663 	cs->s = 1;
2664 	cs->dpl = 0;		/* will be adjusted later */
2665 	cs->p = 1;
2666 	cs->d = 1;
2667 	cs->avl = 0;
2668 
2669 	set_desc_base(ss, 0);	/* flat segment */
2670 	set_desc_limit(ss, 0xfffff);	/* 4GB limit */
2671 	ss->g = 1;		/* 4kb granularity */
2672 	ss->s = 1;
2673 	ss->type = 0x03;	/* Read/Write, Accessed */
2674 	ss->d = 1;		/* 32bit stack segment */
2675 	ss->dpl = 0;
2676 	ss->p = 1;
2677 	ss->l = 0;
2678 	ss->avl = 0;
2679 }
2680 
2681 static bool vendor_intel(struct x86_emulate_ctxt *ctxt)
2682 {
2683 	u32 eax, ebx, ecx, edx;
2684 
2685 	eax = ecx = 0;
2686 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2687 	return is_guest_vendor_intel(ebx, ecx, edx);
2688 }
2689 
2690 static bool em_syscall_is_enabled(struct x86_emulate_ctxt *ctxt)
2691 {
2692 	const struct x86_emulate_ops *ops = ctxt->ops;
2693 	u32 eax, ebx, ecx, edx;
2694 
2695 	/*
2696 	 * syscall should always be enabled in longmode - so only become
2697 	 * vendor specific (cpuid) if other modes are active...
2698 	 */
2699 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2700 		return true;
2701 
2702 	eax = 0x00000000;
2703 	ecx = 0x00000000;
2704 	ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2705 	/*
2706 	 * remark: Intel CPUs only support "syscall" in 64bit longmode. Also a
2707 	 * 64bit guest with a 32bit compat-app running will #UD !! While this
2708 	 * behaviour can be fixed (by emulating) into AMD response - CPUs of
2709 	 * AMD can't behave like Intel.
2710 	 */
2711 	if (is_guest_vendor_intel(ebx, ecx, edx))
2712 		return false;
2713 
2714 	if (is_guest_vendor_amd(ebx, ecx, edx) ||
2715 	    is_guest_vendor_hygon(ebx, ecx, edx))
2716 		return true;
2717 
2718 	/*
2719 	 * default: (not Intel, not AMD, not Hygon), apply Intel's
2720 	 * stricter rules...
2721 	 */
2722 	return false;
2723 }
2724 
2725 static int em_syscall(struct x86_emulate_ctxt *ctxt)
2726 {
2727 	const struct x86_emulate_ops *ops = ctxt->ops;
2728 	struct desc_struct cs, ss;
2729 	u64 msr_data;
2730 	u16 cs_sel, ss_sel;
2731 	u64 efer = 0;
2732 
2733 	/* syscall is not available in real mode */
2734 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2735 	    ctxt->mode == X86EMUL_MODE_VM86)
2736 		return emulate_ud(ctxt);
2737 
2738 	if (!(em_syscall_is_enabled(ctxt)))
2739 		return emulate_ud(ctxt);
2740 
2741 	ops->get_msr(ctxt, MSR_EFER, &efer);
2742 	if (!(efer & EFER_SCE))
2743 		return emulate_ud(ctxt);
2744 
2745 	setup_syscalls_segments(&cs, &ss);
2746 	ops->get_msr(ctxt, MSR_STAR, &msr_data);
2747 	msr_data >>= 32;
2748 	cs_sel = (u16)(msr_data & 0xfffc);
2749 	ss_sel = (u16)(msr_data + 8);
2750 
2751 	if (efer & EFER_LMA) {
2752 		cs.d = 0;
2753 		cs.l = 1;
2754 	}
2755 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2756 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2757 
2758 	*reg_write(ctxt, VCPU_REGS_RCX) = ctxt->_eip;
2759 	if (efer & EFER_LMA) {
2760 #ifdef CONFIG_X86_64
2761 		*reg_write(ctxt, VCPU_REGS_R11) = ctxt->eflags;
2762 
2763 		ops->get_msr(ctxt,
2764 			     ctxt->mode == X86EMUL_MODE_PROT64 ?
2765 			     MSR_LSTAR : MSR_CSTAR, &msr_data);
2766 		ctxt->_eip = msr_data;
2767 
2768 		ops->get_msr(ctxt, MSR_SYSCALL_MASK, &msr_data);
2769 		ctxt->eflags &= ~msr_data;
2770 		ctxt->eflags |= X86_EFLAGS_FIXED;
2771 #endif
2772 	} else {
2773 		/* legacy mode */
2774 		ops->get_msr(ctxt, MSR_STAR, &msr_data);
2775 		ctxt->_eip = (u32)msr_data;
2776 
2777 		ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2778 	}
2779 
2780 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
2781 	return X86EMUL_CONTINUE;
2782 }
2783 
2784 static int em_sysenter(struct x86_emulate_ctxt *ctxt)
2785 {
2786 	const struct x86_emulate_ops *ops = ctxt->ops;
2787 	struct desc_struct cs, ss;
2788 	u64 msr_data;
2789 	u16 cs_sel, ss_sel;
2790 	u64 efer = 0;
2791 
2792 	ops->get_msr(ctxt, MSR_EFER, &efer);
2793 	/* inject #GP if in real mode */
2794 	if (ctxt->mode == X86EMUL_MODE_REAL)
2795 		return emulate_gp(ctxt, 0);
2796 
2797 	/*
2798 	 * Not recognized on AMD in compat mode (but is recognized in legacy
2799 	 * mode).
2800 	 */
2801 	if ((ctxt->mode != X86EMUL_MODE_PROT64) && (efer & EFER_LMA)
2802 	    && !vendor_intel(ctxt))
2803 		return emulate_ud(ctxt);
2804 
2805 	/* sysenter/sysexit have not been tested in 64bit mode. */
2806 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2807 		return X86EMUL_UNHANDLEABLE;
2808 
2809 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2810 	if ((msr_data & 0xfffc) == 0x0)
2811 		return emulate_gp(ctxt, 0);
2812 
2813 	setup_syscalls_segments(&cs, &ss);
2814 	ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2815 	cs_sel = (u16)msr_data & ~SEGMENT_RPL_MASK;
2816 	ss_sel = cs_sel + 8;
2817 	if (efer & EFER_LMA) {
2818 		cs.d = 0;
2819 		cs.l = 1;
2820 	}
2821 
2822 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2823 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2824 
2825 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_EIP, &msr_data);
2826 	ctxt->_eip = (efer & EFER_LMA) ? msr_data : (u32)msr_data;
2827 
2828 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_ESP, &msr_data);
2829 	*reg_write(ctxt, VCPU_REGS_RSP) = (efer & EFER_LMA) ? msr_data :
2830 							      (u32)msr_data;
2831 	if (efer & EFER_LMA)
2832 		ctxt->mode = X86EMUL_MODE_PROT64;
2833 
2834 	return X86EMUL_CONTINUE;
2835 }
2836 
2837 static int em_sysexit(struct x86_emulate_ctxt *ctxt)
2838 {
2839 	const struct x86_emulate_ops *ops = ctxt->ops;
2840 	struct desc_struct cs, ss;
2841 	u64 msr_data, rcx, rdx;
2842 	int usermode;
2843 	u16 cs_sel = 0, ss_sel = 0;
2844 
2845 	/* inject #GP if in real mode or Virtual 8086 mode */
2846 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2847 	    ctxt->mode == X86EMUL_MODE_VM86)
2848 		return emulate_gp(ctxt, 0);
2849 
2850 	setup_syscalls_segments(&cs, &ss);
2851 
2852 	if ((ctxt->rex_prefix & 0x8) != 0x0)
2853 		usermode = X86EMUL_MODE_PROT64;
2854 	else
2855 		usermode = X86EMUL_MODE_PROT32;
2856 
2857 	rcx = reg_read(ctxt, VCPU_REGS_RCX);
2858 	rdx = reg_read(ctxt, VCPU_REGS_RDX);
2859 
2860 	cs.dpl = 3;
2861 	ss.dpl = 3;
2862 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2863 	switch (usermode) {
2864 	case X86EMUL_MODE_PROT32:
2865 		cs_sel = (u16)(msr_data + 16);
2866 		if ((msr_data & 0xfffc) == 0x0)
2867 			return emulate_gp(ctxt, 0);
2868 		ss_sel = (u16)(msr_data + 24);
2869 		rcx = (u32)rcx;
2870 		rdx = (u32)rdx;
2871 		break;
2872 	case X86EMUL_MODE_PROT64:
2873 		cs_sel = (u16)(msr_data + 32);
2874 		if (msr_data == 0x0)
2875 			return emulate_gp(ctxt, 0);
2876 		ss_sel = cs_sel + 8;
2877 		cs.d = 0;
2878 		cs.l = 1;
2879 		if (emul_is_noncanonical_address(rcx, ctxt) ||
2880 		    emul_is_noncanonical_address(rdx, ctxt))
2881 			return emulate_gp(ctxt, 0);
2882 		break;
2883 	}
2884 	cs_sel |= SEGMENT_RPL_MASK;
2885 	ss_sel |= SEGMENT_RPL_MASK;
2886 
2887 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2888 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2889 
2890 	ctxt->_eip = rdx;
2891 	*reg_write(ctxt, VCPU_REGS_RSP) = rcx;
2892 
2893 	return X86EMUL_CONTINUE;
2894 }
2895 
2896 static bool emulator_bad_iopl(struct x86_emulate_ctxt *ctxt)
2897 {
2898 	int iopl;
2899 	if (ctxt->mode == X86EMUL_MODE_REAL)
2900 		return false;
2901 	if (ctxt->mode == X86EMUL_MODE_VM86)
2902 		return true;
2903 	iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
2904 	return ctxt->ops->cpl(ctxt) > iopl;
2905 }
2906 
2907 #define VMWARE_PORT_VMPORT	(0x5658)
2908 #define VMWARE_PORT_VMRPC	(0x5659)
2909 
2910 static bool emulator_io_port_access_allowed(struct x86_emulate_ctxt *ctxt,
2911 					    u16 port, u16 len)
2912 {
2913 	const struct x86_emulate_ops *ops = ctxt->ops;
2914 	struct desc_struct tr_seg;
2915 	u32 base3;
2916 	int r;
2917 	u16 tr, io_bitmap_ptr, perm, bit_idx = port & 0x7;
2918 	unsigned mask = (1 << len) - 1;
2919 	unsigned long base;
2920 
2921 	/*
2922 	 * VMware allows access to these ports even if denied
2923 	 * by TSS I/O permission bitmap. Mimic behavior.
2924 	 */
2925 	if (enable_vmware_backdoor &&
2926 	    ((port == VMWARE_PORT_VMPORT) || (port == VMWARE_PORT_VMRPC)))
2927 		return true;
2928 
2929 	ops->get_segment(ctxt, &tr, &tr_seg, &base3, VCPU_SREG_TR);
2930 	if (!tr_seg.p)
2931 		return false;
2932 	if (desc_limit_scaled(&tr_seg) < 103)
2933 		return false;
2934 	base = get_desc_base(&tr_seg);
2935 #ifdef CONFIG_X86_64
2936 	base |= ((u64)base3) << 32;
2937 #endif
2938 	r = ops->read_std(ctxt, base + 102, &io_bitmap_ptr, 2, NULL, true);
2939 	if (r != X86EMUL_CONTINUE)
2940 		return false;
2941 	if (io_bitmap_ptr + port/8 > desc_limit_scaled(&tr_seg))
2942 		return false;
2943 	r = ops->read_std(ctxt, base + io_bitmap_ptr + port/8, &perm, 2, NULL, true);
2944 	if (r != X86EMUL_CONTINUE)
2945 		return false;
2946 	if ((perm >> bit_idx) & mask)
2947 		return false;
2948 	return true;
2949 }
2950 
2951 static bool emulator_io_permited(struct x86_emulate_ctxt *ctxt,
2952 				 u16 port, u16 len)
2953 {
2954 	if (ctxt->perm_ok)
2955 		return true;
2956 
2957 	if (emulator_bad_iopl(ctxt))
2958 		if (!emulator_io_port_access_allowed(ctxt, port, len))
2959 			return false;
2960 
2961 	ctxt->perm_ok = true;
2962 
2963 	return true;
2964 }
2965 
2966 static void string_registers_quirk(struct x86_emulate_ctxt *ctxt)
2967 {
2968 	/*
2969 	 * Intel CPUs mask the counter and pointers in quite strange
2970 	 * manner when ECX is zero due to REP-string optimizations.
2971 	 */
2972 #ifdef CONFIG_X86_64
2973 	if (ctxt->ad_bytes != 4 || !vendor_intel(ctxt))
2974 		return;
2975 
2976 	*reg_write(ctxt, VCPU_REGS_RCX) = 0;
2977 
2978 	switch (ctxt->b) {
2979 	case 0xa4:	/* movsb */
2980 	case 0xa5:	/* movsd/w */
2981 		*reg_rmw(ctxt, VCPU_REGS_RSI) &= (u32)-1;
2982 		fallthrough;
2983 	case 0xaa:	/* stosb */
2984 	case 0xab:	/* stosd/w */
2985 		*reg_rmw(ctxt, VCPU_REGS_RDI) &= (u32)-1;
2986 	}
2987 #endif
2988 }
2989 
2990 static void save_state_to_tss16(struct x86_emulate_ctxt *ctxt,
2991 				struct tss_segment_16 *tss)
2992 {
2993 	tss->ip = ctxt->_eip;
2994 	tss->flag = ctxt->eflags;
2995 	tss->ax = reg_read(ctxt, VCPU_REGS_RAX);
2996 	tss->cx = reg_read(ctxt, VCPU_REGS_RCX);
2997 	tss->dx = reg_read(ctxt, VCPU_REGS_RDX);
2998 	tss->bx = reg_read(ctxt, VCPU_REGS_RBX);
2999 	tss->sp = reg_read(ctxt, VCPU_REGS_RSP);
3000 	tss->bp = reg_read(ctxt, VCPU_REGS_RBP);
3001 	tss->si = reg_read(ctxt, VCPU_REGS_RSI);
3002 	tss->di = reg_read(ctxt, VCPU_REGS_RDI);
3003 
3004 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
3005 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
3006 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
3007 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
3008 	tss->ldt = get_segment_selector(ctxt, VCPU_SREG_LDTR);
3009 }
3010 
3011 static int load_state_from_tss16(struct x86_emulate_ctxt *ctxt,
3012 				 struct tss_segment_16 *tss)
3013 {
3014 	int ret;
3015 	u8 cpl;
3016 
3017 	ctxt->_eip = tss->ip;
3018 	ctxt->eflags = tss->flag | 2;
3019 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->ax;
3020 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->cx;
3021 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->dx;
3022 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->bx;
3023 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->sp;
3024 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->bp;
3025 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->si;
3026 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->di;
3027 
3028 	/*
3029 	 * SDM says that segment selectors are loaded before segment
3030 	 * descriptors
3031 	 */
3032 	set_segment_selector(ctxt, tss->ldt, VCPU_SREG_LDTR);
3033 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
3034 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
3035 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
3036 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
3037 
3038 	cpl = tss->cs & 3;
3039 
3040 	/*
3041 	 * Now load segment descriptors. If fault happens at this stage
3042 	 * it is handled in a context of new task
3043 	 */
3044 	ret = __load_segment_descriptor(ctxt, tss->ldt, VCPU_SREG_LDTR, cpl,
3045 					X86_TRANSFER_TASK_SWITCH, NULL);
3046 	if (ret != X86EMUL_CONTINUE)
3047 		return ret;
3048 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3049 					X86_TRANSFER_TASK_SWITCH, NULL);
3050 	if (ret != X86EMUL_CONTINUE)
3051 		return ret;
3052 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3053 					X86_TRANSFER_TASK_SWITCH, NULL);
3054 	if (ret != X86EMUL_CONTINUE)
3055 		return ret;
3056 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3057 					X86_TRANSFER_TASK_SWITCH, NULL);
3058 	if (ret != X86EMUL_CONTINUE)
3059 		return ret;
3060 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3061 					X86_TRANSFER_TASK_SWITCH, NULL);
3062 	if (ret != X86EMUL_CONTINUE)
3063 		return ret;
3064 
3065 	return X86EMUL_CONTINUE;
3066 }
3067 
3068 static int task_switch_16(struct x86_emulate_ctxt *ctxt, u16 old_tss_sel,
3069 			  ulong old_tss_base, struct desc_struct *new_desc)
3070 {
3071 	struct tss_segment_16 tss_seg;
3072 	int ret;
3073 	u32 new_tss_base = get_desc_base(new_desc);
3074 
3075 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3076 	if (ret != X86EMUL_CONTINUE)
3077 		return ret;
3078 
3079 	save_state_to_tss16(ctxt, &tss_seg);
3080 
3081 	ret = linear_write_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3082 	if (ret != X86EMUL_CONTINUE)
3083 		return ret;
3084 
3085 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
3086 	if (ret != X86EMUL_CONTINUE)
3087 		return ret;
3088 
3089 	if (old_tss_sel != 0xffff) {
3090 		tss_seg.prev_task_link = old_tss_sel;
3091 
3092 		ret = linear_write_system(ctxt, new_tss_base,
3093 					  &tss_seg.prev_task_link,
3094 					  sizeof(tss_seg.prev_task_link));
3095 		if (ret != X86EMUL_CONTINUE)
3096 			return ret;
3097 	}
3098 
3099 	return load_state_from_tss16(ctxt, &tss_seg);
3100 }
3101 
3102 static void save_state_to_tss32(struct x86_emulate_ctxt *ctxt,
3103 				struct tss_segment_32 *tss)
3104 {
3105 	/* CR3 and ldt selector are not saved intentionally */
3106 	tss->eip = ctxt->_eip;
3107 	tss->eflags = ctxt->eflags;
3108 	tss->eax = reg_read(ctxt, VCPU_REGS_RAX);
3109 	tss->ecx = reg_read(ctxt, VCPU_REGS_RCX);
3110 	tss->edx = reg_read(ctxt, VCPU_REGS_RDX);
3111 	tss->ebx = reg_read(ctxt, VCPU_REGS_RBX);
3112 	tss->esp = reg_read(ctxt, VCPU_REGS_RSP);
3113 	tss->ebp = reg_read(ctxt, VCPU_REGS_RBP);
3114 	tss->esi = reg_read(ctxt, VCPU_REGS_RSI);
3115 	tss->edi = reg_read(ctxt, VCPU_REGS_RDI);
3116 
3117 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
3118 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
3119 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
3120 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
3121 	tss->fs = get_segment_selector(ctxt, VCPU_SREG_FS);
3122 	tss->gs = get_segment_selector(ctxt, VCPU_SREG_GS);
3123 }
3124 
3125 static int load_state_from_tss32(struct x86_emulate_ctxt *ctxt,
3126 				 struct tss_segment_32 *tss)
3127 {
3128 	int ret;
3129 	u8 cpl;
3130 
3131 	if (ctxt->ops->set_cr(ctxt, 3, tss->cr3))
3132 		return emulate_gp(ctxt, 0);
3133 	ctxt->_eip = tss->eip;
3134 	ctxt->eflags = tss->eflags | 2;
3135 
3136 	/* General purpose registers */
3137 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->eax;
3138 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->ecx;
3139 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->edx;
3140 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->ebx;
3141 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->esp;
3142 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->ebp;
3143 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->esi;
3144 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->edi;
3145 
3146 	/*
3147 	 * SDM says that segment selectors are loaded before segment
3148 	 * descriptors.  This is important because CPL checks will
3149 	 * use CS.RPL.
3150 	 */
3151 	set_segment_selector(ctxt, tss->ldt_selector, VCPU_SREG_LDTR);
3152 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
3153 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
3154 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
3155 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
3156 	set_segment_selector(ctxt, tss->fs, VCPU_SREG_FS);
3157 	set_segment_selector(ctxt, tss->gs, VCPU_SREG_GS);
3158 
3159 	/*
3160 	 * If we're switching between Protected Mode and VM86, we need to make
3161 	 * sure to update the mode before loading the segment descriptors so
3162 	 * that the selectors are interpreted correctly.
3163 	 */
3164 	if (ctxt->eflags & X86_EFLAGS_VM) {
3165 		ctxt->mode = X86EMUL_MODE_VM86;
3166 		cpl = 3;
3167 	} else {
3168 		ctxt->mode = X86EMUL_MODE_PROT32;
3169 		cpl = tss->cs & 3;
3170 	}
3171 
3172 	/*
3173 	 * Now load segment descriptors. If fault happens at this stage
3174 	 * it is handled in a context of new task
3175 	 */
3176 	ret = __load_segment_descriptor(ctxt, tss->ldt_selector, VCPU_SREG_LDTR,
3177 					cpl, X86_TRANSFER_TASK_SWITCH, NULL);
3178 	if (ret != X86EMUL_CONTINUE)
3179 		return ret;
3180 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3181 					X86_TRANSFER_TASK_SWITCH, NULL);
3182 	if (ret != X86EMUL_CONTINUE)
3183 		return ret;
3184 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3185 					X86_TRANSFER_TASK_SWITCH, NULL);
3186 	if (ret != X86EMUL_CONTINUE)
3187 		return ret;
3188 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3189 					X86_TRANSFER_TASK_SWITCH, NULL);
3190 	if (ret != X86EMUL_CONTINUE)
3191 		return ret;
3192 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3193 					X86_TRANSFER_TASK_SWITCH, NULL);
3194 	if (ret != X86EMUL_CONTINUE)
3195 		return ret;
3196 	ret = __load_segment_descriptor(ctxt, tss->fs, VCPU_SREG_FS, cpl,
3197 					X86_TRANSFER_TASK_SWITCH, NULL);
3198 	if (ret != X86EMUL_CONTINUE)
3199 		return ret;
3200 	ret = __load_segment_descriptor(ctxt, tss->gs, VCPU_SREG_GS, cpl,
3201 					X86_TRANSFER_TASK_SWITCH, NULL);
3202 
3203 	return ret;
3204 }
3205 
3206 static int task_switch_32(struct x86_emulate_ctxt *ctxt, u16 old_tss_sel,
3207 			  ulong old_tss_base, struct desc_struct *new_desc)
3208 {
3209 	struct tss_segment_32 tss_seg;
3210 	int ret;
3211 	u32 new_tss_base = get_desc_base(new_desc);
3212 	u32 eip_offset = offsetof(struct tss_segment_32, eip);
3213 	u32 ldt_sel_offset = offsetof(struct tss_segment_32, ldt_selector);
3214 
3215 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3216 	if (ret != X86EMUL_CONTINUE)
3217 		return ret;
3218 
3219 	save_state_to_tss32(ctxt, &tss_seg);
3220 
3221 	/* Only GP registers and segment selectors are saved */
3222 	ret = linear_write_system(ctxt, old_tss_base + eip_offset, &tss_seg.eip,
3223 				  ldt_sel_offset - eip_offset);
3224 	if (ret != X86EMUL_CONTINUE)
3225 		return ret;
3226 
3227 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
3228 	if (ret != X86EMUL_CONTINUE)
3229 		return ret;
3230 
3231 	if (old_tss_sel != 0xffff) {
3232 		tss_seg.prev_task_link = old_tss_sel;
3233 
3234 		ret = linear_write_system(ctxt, new_tss_base,
3235 					  &tss_seg.prev_task_link,
3236 					  sizeof(tss_seg.prev_task_link));
3237 		if (ret != X86EMUL_CONTINUE)
3238 			return ret;
3239 	}
3240 
3241 	return load_state_from_tss32(ctxt, &tss_seg);
3242 }
3243 
3244 static int emulator_do_task_switch(struct x86_emulate_ctxt *ctxt,
3245 				   u16 tss_selector, int idt_index, int reason,
3246 				   bool has_error_code, u32 error_code)
3247 {
3248 	const struct x86_emulate_ops *ops = ctxt->ops;
3249 	struct desc_struct curr_tss_desc, next_tss_desc;
3250 	int ret;
3251 	u16 old_tss_sel = get_segment_selector(ctxt, VCPU_SREG_TR);
3252 	ulong old_tss_base =
3253 		ops->get_cached_segment_base(ctxt, VCPU_SREG_TR);
3254 	u32 desc_limit;
3255 	ulong desc_addr, dr7;
3256 
3257 	/* FIXME: old_tss_base == ~0 ? */
3258 
3259 	ret = read_segment_descriptor(ctxt, tss_selector, &next_tss_desc, &desc_addr);
3260 	if (ret != X86EMUL_CONTINUE)
3261 		return ret;
3262 	ret = read_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc, &desc_addr);
3263 	if (ret != X86EMUL_CONTINUE)
3264 		return ret;
3265 
3266 	/* FIXME: check that next_tss_desc is tss */
3267 
3268 	/*
3269 	 * Check privileges. The three cases are task switch caused by...
3270 	 *
3271 	 * 1. jmp/call/int to task gate: Check against DPL of the task gate
3272 	 * 2. Exception/IRQ/iret: No check is performed
3273 	 * 3. jmp/call to TSS/task-gate: No check is performed since the
3274 	 *    hardware checks it before exiting.
3275 	 */
3276 	if (reason == TASK_SWITCH_GATE) {
3277 		if (idt_index != -1) {
3278 			/* Software interrupts */
3279 			struct desc_struct task_gate_desc;
3280 			int dpl;
3281 
3282 			ret = read_interrupt_descriptor(ctxt, idt_index,
3283 							&task_gate_desc);
3284 			if (ret != X86EMUL_CONTINUE)
3285 				return ret;
3286 
3287 			dpl = task_gate_desc.dpl;
3288 			if ((tss_selector & 3) > dpl || ops->cpl(ctxt) > dpl)
3289 				return emulate_gp(ctxt, (idt_index << 3) | 0x2);
3290 		}
3291 	}
3292 
3293 	desc_limit = desc_limit_scaled(&next_tss_desc);
3294 	if (!next_tss_desc.p ||
3295 	    ((desc_limit < 0x67 && (next_tss_desc.type & 8)) ||
3296 	     desc_limit < 0x2b)) {
3297 		return emulate_ts(ctxt, tss_selector & 0xfffc);
3298 	}
3299 
3300 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3301 		curr_tss_desc.type &= ~(1 << 1); /* clear busy flag */
3302 		write_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc);
3303 	}
3304 
3305 	if (reason == TASK_SWITCH_IRET)
3306 		ctxt->eflags = ctxt->eflags & ~X86_EFLAGS_NT;
3307 
3308 	/* set back link to prev task only if NT bit is set in eflags
3309 	   note that old_tss_sel is not used after this point */
3310 	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
3311 		old_tss_sel = 0xffff;
3312 
3313 	if (next_tss_desc.type & 8)
3314 		ret = task_switch_32(ctxt, old_tss_sel, old_tss_base, &next_tss_desc);
3315 	else
3316 		ret = task_switch_16(ctxt, old_tss_sel,
3317 				     old_tss_base, &next_tss_desc);
3318 	if (ret != X86EMUL_CONTINUE)
3319 		return ret;
3320 
3321 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE)
3322 		ctxt->eflags = ctxt->eflags | X86_EFLAGS_NT;
3323 
3324 	if (reason != TASK_SWITCH_IRET) {
3325 		next_tss_desc.type |= (1 << 1); /* set busy flag */
3326 		write_segment_descriptor(ctxt, tss_selector, &next_tss_desc);
3327 	}
3328 
3329 	ops->set_cr(ctxt, 0,  ops->get_cr(ctxt, 0) | X86_CR0_TS);
3330 	ops->set_segment(ctxt, tss_selector, &next_tss_desc, 0, VCPU_SREG_TR);
3331 
3332 	if (has_error_code) {
3333 		ctxt->op_bytes = ctxt->ad_bytes = (next_tss_desc.type & 8) ? 4 : 2;
3334 		ctxt->lock_prefix = 0;
3335 		ctxt->src.val = (unsigned long) error_code;
3336 		ret = em_push(ctxt);
3337 	}
3338 
3339 	ops->get_dr(ctxt, 7, &dr7);
3340 	ops->set_dr(ctxt, 7, dr7 & ~(DR_LOCAL_ENABLE_MASK | DR_LOCAL_SLOWDOWN));
3341 
3342 	return ret;
3343 }
3344 
3345 int emulator_task_switch(struct x86_emulate_ctxt *ctxt,
3346 			 u16 tss_selector, int idt_index, int reason,
3347 			 bool has_error_code, u32 error_code)
3348 {
3349 	int rc;
3350 
3351 	invalidate_registers(ctxt);
3352 	ctxt->_eip = ctxt->eip;
3353 	ctxt->dst.type = OP_NONE;
3354 
3355 	rc = emulator_do_task_switch(ctxt, tss_selector, idt_index, reason,
3356 				     has_error_code, error_code);
3357 
3358 	if (rc == X86EMUL_CONTINUE) {
3359 		ctxt->eip = ctxt->_eip;
3360 		writeback_registers(ctxt);
3361 	}
3362 
3363 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
3364 }
3365 
3366 static void string_addr_inc(struct x86_emulate_ctxt *ctxt, int reg,
3367 		struct operand *op)
3368 {
3369 	int df = (ctxt->eflags & X86_EFLAGS_DF) ? -op->count : op->count;
3370 
3371 	register_address_increment(ctxt, reg, df * op->bytes);
3372 	op->addr.mem.ea = register_address(ctxt, reg);
3373 }
3374 
3375 static int em_das(struct x86_emulate_ctxt *ctxt)
3376 {
3377 	u8 al, old_al;
3378 	bool af, cf, old_cf;
3379 
3380 	cf = ctxt->eflags & X86_EFLAGS_CF;
3381 	al = ctxt->dst.val;
3382 
3383 	old_al = al;
3384 	old_cf = cf;
3385 	cf = false;
3386 	af = ctxt->eflags & X86_EFLAGS_AF;
3387 	if ((al & 0x0f) > 9 || af) {
3388 		al -= 6;
3389 		cf = old_cf | (al >= 250);
3390 		af = true;
3391 	} else {
3392 		af = false;
3393 	}
3394 	if (old_al > 0x99 || old_cf) {
3395 		al -= 0x60;
3396 		cf = true;
3397 	}
3398 
3399 	ctxt->dst.val = al;
3400 	/* Set PF, ZF, SF */
3401 	ctxt->src.type = OP_IMM;
3402 	ctxt->src.val = 0;
3403 	ctxt->src.bytes = 1;
3404 	fastop(ctxt, em_or);
3405 	ctxt->eflags &= ~(X86_EFLAGS_AF | X86_EFLAGS_CF);
3406 	if (cf)
3407 		ctxt->eflags |= X86_EFLAGS_CF;
3408 	if (af)
3409 		ctxt->eflags |= X86_EFLAGS_AF;
3410 	return X86EMUL_CONTINUE;
3411 }
3412 
3413 static int em_aam(struct x86_emulate_ctxt *ctxt)
3414 {
3415 	u8 al, ah;
3416 
3417 	if (ctxt->src.val == 0)
3418 		return emulate_de(ctxt);
3419 
3420 	al = ctxt->dst.val & 0xff;
3421 	ah = al / ctxt->src.val;
3422 	al %= ctxt->src.val;
3423 
3424 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al | (ah << 8);
3425 
3426 	/* Set PF, ZF, SF */
3427 	ctxt->src.type = OP_IMM;
3428 	ctxt->src.val = 0;
3429 	ctxt->src.bytes = 1;
3430 	fastop(ctxt, em_or);
3431 
3432 	return X86EMUL_CONTINUE;
3433 }
3434 
3435 static int em_aad(struct x86_emulate_ctxt *ctxt)
3436 {
3437 	u8 al = ctxt->dst.val & 0xff;
3438 	u8 ah = (ctxt->dst.val >> 8) & 0xff;
3439 
3440 	al = (al + (ah * ctxt->src.val)) & 0xff;
3441 
3442 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al;
3443 
3444 	/* Set PF, ZF, SF */
3445 	ctxt->src.type = OP_IMM;
3446 	ctxt->src.val = 0;
3447 	ctxt->src.bytes = 1;
3448 	fastop(ctxt, em_or);
3449 
3450 	return X86EMUL_CONTINUE;
3451 }
3452 
3453 static int em_call(struct x86_emulate_ctxt *ctxt)
3454 {
3455 	int rc;
3456 	long rel = ctxt->src.val;
3457 
3458 	ctxt->src.val = (unsigned long)ctxt->_eip;
3459 	rc = jmp_rel(ctxt, rel);
3460 	if (rc != X86EMUL_CONTINUE)
3461 		return rc;
3462 	return em_push(ctxt);
3463 }
3464 
3465 static int em_call_far(struct x86_emulate_ctxt *ctxt)
3466 {
3467 	u16 sel, old_cs;
3468 	ulong old_eip;
3469 	int rc;
3470 	struct desc_struct old_desc, new_desc;
3471 	const struct x86_emulate_ops *ops = ctxt->ops;
3472 	int cpl = ctxt->ops->cpl(ctxt);
3473 	enum x86emul_mode prev_mode = ctxt->mode;
3474 
3475 	old_eip = ctxt->_eip;
3476 	ops->get_segment(ctxt, &old_cs, &old_desc, NULL, VCPU_SREG_CS);
3477 
3478 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
3479 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
3480 				       X86_TRANSFER_CALL_JMP, &new_desc);
3481 	if (rc != X86EMUL_CONTINUE)
3482 		return rc;
3483 
3484 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
3485 	if (rc != X86EMUL_CONTINUE)
3486 		goto fail;
3487 
3488 	ctxt->src.val = old_cs;
3489 	rc = em_push(ctxt);
3490 	if (rc != X86EMUL_CONTINUE)
3491 		goto fail;
3492 
3493 	ctxt->src.val = old_eip;
3494 	rc = em_push(ctxt);
3495 	/* If we failed, we tainted the memory, but the very least we should
3496 	   restore cs */
3497 	if (rc != X86EMUL_CONTINUE) {
3498 		pr_warn_once("faulting far call emulation tainted memory\n");
3499 		goto fail;
3500 	}
3501 	return rc;
3502 fail:
3503 	ops->set_segment(ctxt, old_cs, &old_desc, 0, VCPU_SREG_CS);
3504 	ctxt->mode = prev_mode;
3505 	return rc;
3506 
3507 }
3508 
3509 static int em_ret_near_imm(struct x86_emulate_ctxt *ctxt)
3510 {
3511 	int rc;
3512 	unsigned long eip;
3513 
3514 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
3515 	if (rc != X86EMUL_CONTINUE)
3516 		return rc;
3517 	rc = assign_eip_near(ctxt, eip);
3518 	if (rc != X86EMUL_CONTINUE)
3519 		return rc;
3520 	rsp_increment(ctxt, ctxt->src.val);
3521 	return X86EMUL_CONTINUE;
3522 }
3523 
3524 static int em_xchg(struct x86_emulate_ctxt *ctxt)
3525 {
3526 	/* Write back the register source. */
3527 	ctxt->src.val = ctxt->dst.val;
3528 	write_register_operand(&ctxt->src);
3529 
3530 	/* Write back the memory destination with implicit LOCK prefix. */
3531 	ctxt->dst.val = ctxt->src.orig_val;
3532 	ctxt->lock_prefix = 1;
3533 	return X86EMUL_CONTINUE;
3534 }
3535 
3536 static int em_imul_3op(struct x86_emulate_ctxt *ctxt)
3537 {
3538 	ctxt->dst.val = ctxt->src2.val;
3539 	return fastop(ctxt, em_imul);
3540 }
3541 
3542 static int em_cwd(struct x86_emulate_ctxt *ctxt)
3543 {
3544 	ctxt->dst.type = OP_REG;
3545 	ctxt->dst.bytes = ctxt->src.bytes;
3546 	ctxt->dst.addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
3547 	ctxt->dst.val = ~((ctxt->src.val >> (ctxt->src.bytes * 8 - 1)) - 1);
3548 
3549 	return X86EMUL_CONTINUE;
3550 }
3551 
3552 static int em_rdpid(struct x86_emulate_ctxt *ctxt)
3553 {
3554 	u64 tsc_aux = 0;
3555 
3556 	if (!ctxt->ops->guest_has_rdpid(ctxt))
3557 		return emulate_ud(ctxt);
3558 
3559 	ctxt->ops->get_msr(ctxt, MSR_TSC_AUX, &tsc_aux);
3560 	ctxt->dst.val = tsc_aux;
3561 	return X86EMUL_CONTINUE;
3562 }
3563 
3564 static int em_rdtsc(struct x86_emulate_ctxt *ctxt)
3565 {
3566 	u64 tsc = 0;
3567 
3568 	ctxt->ops->get_msr(ctxt, MSR_IA32_TSC, &tsc);
3569 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)tsc;
3570 	*reg_write(ctxt, VCPU_REGS_RDX) = tsc >> 32;
3571 	return X86EMUL_CONTINUE;
3572 }
3573 
3574 static int em_rdpmc(struct x86_emulate_ctxt *ctxt)
3575 {
3576 	u64 pmc;
3577 
3578 	if (ctxt->ops->read_pmc(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &pmc))
3579 		return emulate_gp(ctxt, 0);
3580 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)pmc;
3581 	*reg_write(ctxt, VCPU_REGS_RDX) = pmc >> 32;
3582 	return X86EMUL_CONTINUE;
3583 }
3584 
3585 static int em_mov(struct x86_emulate_ctxt *ctxt)
3586 {
3587 	memcpy(ctxt->dst.valptr, ctxt->src.valptr, sizeof(ctxt->src.valptr));
3588 	return X86EMUL_CONTINUE;
3589 }
3590 
3591 static int em_movbe(struct x86_emulate_ctxt *ctxt)
3592 {
3593 	u16 tmp;
3594 
3595 	if (!ctxt->ops->guest_has_movbe(ctxt))
3596 		return emulate_ud(ctxt);
3597 
3598 	switch (ctxt->op_bytes) {
3599 	case 2:
3600 		/*
3601 		 * From MOVBE definition: "...When the operand size is 16 bits,
3602 		 * the upper word of the destination register remains unchanged
3603 		 * ..."
3604 		 *
3605 		 * Both casting ->valptr and ->val to u16 breaks strict aliasing
3606 		 * rules so we have to do the operation almost per hand.
3607 		 */
3608 		tmp = (u16)ctxt->src.val;
3609 		ctxt->dst.val &= ~0xffffUL;
3610 		ctxt->dst.val |= (unsigned long)swab16(tmp);
3611 		break;
3612 	case 4:
3613 		ctxt->dst.val = swab32((u32)ctxt->src.val);
3614 		break;
3615 	case 8:
3616 		ctxt->dst.val = swab64(ctxt->src.val);
3617 		break;
3618 	default:
3619 		BUG();
3620 	}
3621 	return X86EMUL_CONTINUE;
3622 }
3623 
3624 static int em_cr_write(struct x86_emulate_ctxt *ctxt)
3625 {
3626 	if (ctxt->ops->set_cr(ctxt, ctxt->modrm_reg, ctxt->src.val))
3627 		return emulate_gp(ctxt, 0);
3628 
3629 	/* Disable writeback. */
3630 	ctxt->dst.type = OP_NONE;
3631 	return X86EMUL_CONTINUE;
3632 }
3633 
3634 static int em_dr_write(struct x86_emulate_ctxt *ctxt)
3635 {
3636 	unsigned long val;
3637 
3638 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3639 		val = ctxt->src.val & ~0ULL;
3640 	else
3641 		val = ctxt->src.val & ~0U;
3642 
3643 	/* #UD condition is already handled. */
3644 	if (ctxt->ops->set_dr(ctxt, ctxt->modrm_reg, val) < 0)
3645 		return emulate_gp(ctxt, 0);
3646 
3647 	/* Disable writeback. */
3648 	ctxt->dst.type = OP_NONE;
3649 	return X86EMUL_CONTINUE;
3650 }
3651 
3652 static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
3653 {
3654 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3655 	u64 msr_data;
3656 	int r;
3657 
3658 	msr_data = (u32)reg_read(ctxt, VCPU_REGS_RAX)
3659 		| ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32);
3660 	r = ctxt->ops->set_msr_with_filter(ctxt, msr_index, msr_data);
3661 
3662 	if (r == X86EMUL_IO_NEEDED)
3663 		return r;
3664 
3665 	if (r > 0)
3666 		return emulate_gp(ctxt, 0);
3667 
3668 	return r < 0 ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
3669 }
3670 
3671 static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
3672 {
3673 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3674 	u64 msr_data;
3675 	int r;
3676 
3677 	r = ctxt->ops->get_msr_with_filter(ctxt, msr_index, &msr_data);
3678 
3679 	if (r == X86EMUL_IO_NEEDED)
3680 		return r;
3681 
3682 	if (r)
3683 		return emulate_gp(ctxt, 0);
3684 
3685 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
3686 	*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
3687 	return X86EMUL_CONTINUE;
3688 }
3689 
3690 static int em_store_sreg(struct x86_emulate_ctxt *ctxt, int segment)
3691 {
3692 	if (segment > VCPU_SREG_GS &&
3693 	    (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3694 	    ctxt->ops->cpl(ctxt) > 0)
3695 		return emulate_gp(ctxt, 0);
3696 
3697 	ctxt->dst.val = get_segment_selector(ctxt, segment);
3698 	if (ctxt->dst.bytes == 4 && ctxt->dst.type == OP_MEM)
3699 		ctxt->dst.bytes = 2;
3700 	return X86EMUL_CONTINUE;
3701 }
3702 
3703 static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt)
3704 {
3705 	if (ctxt->modrm_reg > VCPU_SREG_GS)
3706 		return emulate_ud(ctxt);
3707 
3708 	return em_store_sreg(ctxt, ctxt->modrm_reg);
3709 }
3710 
3711 static int em_mov_sreg_rm(struct x86_emulate_ctxt *ctxt)
3712 {
3713 	u16 sel = ctxt->src.val;
3714 
3715 	if (ctxt->modrm_reg == VCPU_SREG_CS || ctxt->modrm_reg > VCPU_SREG_GS)
3716 		return emulate_ud(ctxt);
3717 
3718 	if (ctxt->modrm_reg == VCPU_SREG_SS)
3719 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
3720 
3721 	/* Disable writeback. */
3722 	ctxt->dst.type = OP_NONE;
3723 	return load_segment_descriptor(ctxt, sel, ctxt->modrm_reg);
3724 }
3725 
3726 static int em_sldt(struct x86_emulate_ctxt *ctxt)
3727 {
3728 	return em_store_sreg(ctxt, VCPU_SREG_LDTR);
3729 }
3730 
3731 static int em_lldt(struct x86_emulate_ctxt *ctxt)
3732 {
3733 	u16 sel = ctxt->src.val;
3734 
3735 	/* Disable writeback. */
3736 	ctxt->dst.type = OP_NONE;
3737 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_LDTR);
3738 }
3739 
3740 static int em_str(struct x86_emulate_ctxt *ctxt)
3741 {
3742 	return em_store_sreg(ctxt, VCPU_SREG_TR);
3743 }
3744 
3745 static int em_ltr(struct x86_emulate_ctxt *ctxt)
3746 {
3747 	u16 sel = ctxt->src.val;
3748 
3749 	/* Disable writeback. */
3750 	ctxt->dst.type = OP_NONE;
3751 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_TR);
3752 }
3753 
3754 static int em_invlpg(struct x86_emulate_ctxt *ctxt)
3755 {
3756 	int rc;
3757 	ulong linear;
3758 
3759 	rc = linearize(ctxt, ctxt->src.addr.mem, 1, false, &linear);
3760 	if (rc == X86EMUL_CONTINUE)
3761 		ctxt->ops->invlpg(ctxt, linear);
3762 	/* Disable writeback. */
3763 	ctxt->dst.type = OP_NONE;
3764 	return X86EMUL_CONTINUE;
3765 }
3766 
3767 static int em_clts(struct x86_emulate_ctxt *ctxt)
3768 {
3769 	ulong cr0;
3770 
3771 	cr0 = ctxt->ops->get_cr(ctxt, 0);
3772 	cr0 &= ~X86_CR0_TS;
3773 	ctxt->ops->set_cr(ctxt, 0, cr0);
3774 	return X86EMUL_CONTINUE;
3775 }
3776 
3777 static int em_hypercall(struct x86_emulate_ctxt *ctxt)
3778 {
3779 	int rc = ctxt->ops->fix_hypercall(ctxt);
3780 
3781 	if (rc != X86EMUL_CONTINUE)
3782 		return rc;
3783 
3784 	/* Let the processor re-execute the fixed hypercall */
3785 	ctxt->_eip = ctxt->eip;
3786 	/* Disable writeback. */
3787 	ctxt->dst.type = OP_NONE;
3788 	return X86EMUL_CONTINUE;
3789 }
3790 
3791 static int emulate_store_desc_ptr(struct x86_emulate_ctxt *ctxt,
3792 				  void (*get)(struct x86_emulate_ctxt *ctxt,
3793 					      struct desc_ptr *ptr))
3794 {
3795 	struct desc_ptr desc_ptr;
3796 
3797 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3798 	    ctxt->ops->cpl(ctxt) > 0)
3799 		return emulate_gp(ctxt, 0);
3800 
3801 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3802 		ctxt->op_bytes = 8;
3803 	get(ctxt, &desc_ptr);
3804 	if (ctxt->op_bytes == 2) {
3805 		ctxt->op_bytes = 4;
3806 		desc_ptr.address &= 0x00ffffff;
3807 	}
3808 	/* Disable writeback. */
3809 	ctxt->dst.type = OP_NONE;
3810 	return segmented_write_std(ctxt, ctxt->dst.addr.mem,
3811 				   &desc_ptr, 2 + ctxt->op_bytes);
3812 }
3813 
3814 static int em_sgdt(struct x86_emulate_ctxt *ctxt)
3815 {
3816 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_gdt);
3817 }
3818 
3819 static int em_sidt(struct x86_emulate_ctxt *ctxt)
3820 {
3821 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_idt);
3822 }
3823 
3824 static int em_lgdt_lidt(struct x86_emulate_ctxt *ctxt, bool lgdt)
3825 {
3826 	struct desc_ptr desc_ptr;
3827 	int rc;
3828 
3829 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3830 		ctxt->op_bytes = 8;
3831 	rc = read_descriptor(ctxt, ctxt->src.addr.mem,
3832 			     &desc_ptr.size, &desc_ptr.address,
3833 			     ctxt->op_bytes);
3834 	if (rc != X86EMUL_CONTINUE)
3835 		return rc;
3836 	if (ctxt->mode == X86EMUL_MODE_PROT64 &&
3837 	    emul_is_noncanonical_address(desc_ptr.address, ctxt))
3838 		return emulate_gp(ctxt, 0);
3839 	if (lgdt)
3840 		ctxt->ops->set_gdt(ctxt, &desc_ptr);
3841 	else
3842 		ctxt->ops->set_idt(ctxt, &desc_ptr);
3843 	/* Disable writeback. */
3844 	ctxt->dst.type = OP_NONE;
3845 	return X86EMUL_CONTINUE;
3846 }
3847 
3848 static int em_lgdt(struct x86_emulate_ctxt *ctxt)
3849 {
3850 	return em_lgdt_lidt(ctxt, true);
3851 }
3852 
3853 static int em_lidt(struct x86_emulate_ctxt *ctxt)
3854 {
3855 	return em_lgdt_lidt(ctxt, false);
3856 }
3857 
3858 static int em_smsw(struct x86_emulate_ctxt *ctxt)
3859 {
3860 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3861 	    ctxt->ops->cpl(ctxt) > 0)
3862 		return emulate_gp(ctxt, 0);
3863 
3864 	if (ctxt->dst.type == OP_MEM)
3865 		ctxt->dst.bytes = 2;
3866 	ctxt->dst.val = ctxt->ops->get_cr(ctxt, 0);
3867 	return X86EMUL_CONTINUE;
3868 }
3869 
3870 static int em_lmsw(struct x86_emulate_ctxt *ctxt)
3871 {
3872 	ctxt->ops->set_cr(ctxt, 0, (ctxt->ops->get_cr(ctxt, 0) & ~0x0eul)
3873 			  | (ctxt->src.val & 0x0f));
3874 	ctxt->dst.type = OP_NONE;
3875 	return X86EMUL_CONTINUE;
3876 }
3877 
3878 static int em_loop(struct x86_emulate_ctxt *ctxt)
3879 {
3880 	int rc = X86EMUL_CONTINUE;
3881 
3882 	register_address_increment(ctxt, VCPU_REGS_RCX, -1);
3883 	if ((address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) != 0) &&
3884 	    (ctxt->b == 0xe2 || test_cc(ctxt->b ^ 0x5, ctxt->eflags)))
3885 		rc = jmp_rel(ctxt, ctxt->src.val);
3886 
3887 	return rc;
3888 }
3889 
3890 static int em_jcxz(struct x86_emulate_ctxt *ctxt)
3891 {
3892 	int rc = X86EMUL_CONTINUE;
3893 
3894 	if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0)
3895 		rc = jmp_rel(ctxt, ctxt->src.val);
3896 
3897 	return rc;
3898 }
3899 
3900 static int em_in(struct x86_emulate_ctxt *ctxt)
3901 {
3902 	if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val,
3903 			     &ctxt->dst.val))
3904 		return X86EMUL_IO_NEEDED;
3905 
3906 	return X86EMUL_CONTINUE;
3907 }
3908 
3909 static int em_out(struct x86_emulate_ctxt *ctxt)
3910 {
3911 	ctxt->ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val,
3912 				    &ctxt->src.val, 1);
3913 	/* Disable writeback. */
3914 	ctxt->dst.type = OP_NONE;
3915 	return X86EMUL_CONTINUE;
3916 }
3917 
3918 static int em_cli(struct x86_emulate_ctxt *ctxt)
3919 {
3920 	if (emulator_bad_iopl(ctxt))
3921 		return emulate_gp(ctxt, 0);
3922 
3923 	ctxt->eflags &= ~X86_EFLAGS_IF;
3924 	return X86EMUL_CONTINUE;
3925 }
3926 
3927 static int em_sti(struct x86_emulate_ctxt *ctxt)
3928 {
3929 	if (emulator_bad_iopl(ctxt))
3930 		return emulate_gp(ctxt, 0);
3931 
3932 	ctxt->interruptibility = KVM_X86_SHADOW_INT_STI;
3933 	ctxt->eflags |= X86_EFLAGS_IF;
3934 	return X86EMUL_CONTINUE;
3935 }
3936 
3937 static int em_cpuid(struct x86_emulate_ctxt *ctxt)
3938 {
3939 	u32 eax, ebx, ecx, edx;
3940 	u64 msr = 0;
3941 
3942 	ctxt->ops->get_msr(ctxt, MSR_MISC_FEATURES_ENABLES, &msr);
3943 	if (msr & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3944 	    ctxt->ops->cpl(ctxt)) {
3945 		return emulate_gp(ctxt, 0);
3946 	}
3947 
3948 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3949 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3950 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
3951 	*reg_write(ctxt, VCPU_REGS_RAX) = eax;
3952 	*reg_write(ctxt, VCPU_REGS_RBX) = ebx;
3953 	*reg_write(ctxt, VCPU_REGS_RCX) = ecx;
3954 	*reg_write(ctxt, VCPU_REGS_RDX) = edx;
3955 	return X86EMUL_CONTINUE;
3956 }
3957 
3958 static int em_sahf(struct x86_emulate_ctxt *ctxt)
3959 {
3960 	u32 flags;
3961 
3962 	flags = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
3963 		X86_EFLAGS_SF;
3964 	flags &= *reg_rmw(ctxt, VCPU_REGS_RAX) >> 8;
3965 
3966 	ctxt->eflags &= ~0xffUL;
3967 	ctxt->eflags |= flags | X86_EFLAGS_FIXED;
3968 	return X86EMUL_CONTINUE;
3969 }
3970 
3971 static int em_lahf(struct x86_emulate_ctxt *ctxt)
3972 {
3973 	*reg_rmw(ctxt, VCPU_REGS_RAX) &= ~0xff00UL;
3974 	*reg_rmw(ctxt, VCPU_REGS_RAX) |= (ctxt->eflags & 0xff) << 8;
3975 	return X86EMUL_CONTINUE;
3976 }
3977 
3978 static int em_bswap(struct x86_emulate_ctxt *ctxt)
3979 {
3980 	switch (ctxt->op_bytes) {
3981 #ifdef CONFIG_X86_64
3982 	case 8:
3983 		asm("bswap %0" : "+r"(ctxt->dst.val));
3984 		break;
3985 #endif
3986 	default:
3987 		asm("bswap %0" : "+r"(*(u32 *)&ctxt->dst.val));
3988 		break;
3989 	}
3990 	return X86EMUL_CONTINUE;
3991 }
3992 
3993 static int em_clflush(struct x86_emulate_ctxt *ctxt)
3994 {
3995 	/* emulating clflush regardless of cpuid */
3996 	return X86EMUL_CONTINUE;
3997 }
3998 
3999 static int em_clflushopt(struct x86_emulate_ctxt *ctxt)
4000 {
4001 	/* emulating clflushopt regardless of cpuid */
4002 	return X86EMUL_CONTINUE;
4003 }
4004 
4005 static int em_movsxd(struct x86_emulate_ctxt *ctxt)
4006 {
4007 	ctxt->dst.val = (s32) ctxt->src.val;
4008 	return X86EMUL_CONTINUE;
4009 }
4010 
4011 static int check_fxsr(struct x86_emulate_ctxt *ctxt)
4012 {
4013 	if (!ctxt->ops->guest_has_fxsr(ctxt))
4014 		return emulate_ud(ctxt);
4015 
4016 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
4017 		return emulate_nm(ctxt);
4018 
4019 	/*
4020 	 * Don't emulate a case that should never be hit, instead of working
4021 	 * around a lack of fxsave64/fxrstor64 on old compilers.
4022 	 */
4023 	if (ctxt->mode >= X86EMUL_MODE_PROT64)
4024 		return X86EMUL_UNHANDLEABLE;
4025 
4026 	return X86EMUL_CONTINUE;
4027 }
4028 
4029 /*
4030  * Hardware doesn't save and restore XMM 0-7 without CR4.OSFXSR, but does save
4031  * and restore MXCSR.
4032  */
4033 static size_t __fxstate_size(int nregs)
4034 {
4035 	return offsetof(struct fxregs_state, xmm_space[0]) + nregs * 16;
4036 }
4037 
4038 static inline size_t fxstate_size(struct x86_emulate_ctxt *ctxt)
4039 {
4040 	bool cr4_osfxsr;
4041 	if (ctxt->mode == X86EMUL_MODE_PROT64)
4042 		return __fxstate_size(16);
4043 
4044 	cr4_osfxsr = ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR;
4045 	return __fxstate_size(cr4_osfxsr ? 8 : 0);
4046 }
4047 
4048 /*
4049  * FXSAVE and FXRSTOR have 4 different formats depending on execution mode,
4050  *  1) 16 bit mode
4051  *  2) 32 bit mode
4052  *     - like (1), but FIP and FDP (foo) are only 16 bit.  At least Intel CPUs
4053  *       preserve whole 32 bit values, though, so (1) and (2) are the same wrt.
4054  *       save and restore
4055  *  3) 64-bit mode with REX.W prefix
4056  *     - like (2), but XMM 8-15 are being saved and restored
4057  *  4) 64-bit mode without REX.W prefix
4058  *     - like (3), but FIP and FDP are 64 bit
4059  *
4060  * Emulation uses (3) for (1) and (2) and preserves XMM 8-15 to reach the
4061  * desired result.  (4) is not emulated.
4062  *
4063  * Note: Guest and host CPUID.(EAX=07H,ECX=0H):EBX[bit 13] (deprecate FPU CS
4064  * and FPU DS) should match.
4065  */
4066 static int em_fxsave(struct x86_emulate_ctxt *ctxt)
4067 {
4068 	struct fxregs_state fx_state;
4069 	int rc;
4070 
4071 	rc = check_fxsr(ctxt);
4072 	if (rc != X86EMUL_CONTINUE)
4073 		return rc;
4074 
4075 	kvm_fpu_get();
4076 
4077 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_state));
4078 
4079 	kvm_fpu_put();
4080 
4081 	if (rc != X86EMUL_CONTINUE)
4082 		return rc;
4083 
4084 	return segmented_write_std(ctxt, ctxt->memop.addr.mem, &fx_state,
4085 		                   fxstate_size(ctxt));
4086 }
4087 
4088 /*
4089  * FXRSTOR might restore XMM registers not provided by the guest. Fill
4090  * in the host registers (via FXSAVE) instead, so they won't be modified.
4091  * (preemption has to stay disabled until FXRSTOR).
4092  *
4093  * Use noinline to keep the stack for other functions called by callers small.
4094  */
4095 static noinline int fxregs_fixup(struct fxregs_state *fx_state,
4096 				 const size_t used_size)
4097 {
4098 	struct fxregs_state fx_tmp;
4099 	int rc;
4100 
4101 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_tmp));
4102 	memcpy((void *)fx_state + used_size, (void *)&fx_tmp + used_size,
4103 	       __fxstate_size(16) - used_size);
4104 
4105 	return rc;
4106 }
4107 
4108 static int em_fxrstor(struct x86_emulate_ctxt *ctxt)
4109 {
4110 	struct fxregs_state fx_state;
4111 	int rc;
4112 	size_t size;
4113 
4114 	rc = check_fxsr(ctxt);
4115 	if (rc != X86EMUL_CONTINUE)
4116 		return rc;
4117 
4118 	size = fxstate_size(ctxt);
4119 	rc = segmented_read_std(ctxt, ctxt->memop.addr.mem, &fx_state, size);
4120 	if (rc != X86EMUL_CONTINUE)
4121 		return rc;
4122 
4123 	kvm_fpu_get();
4124 
4125 	if (size < __fxstate_size(16)) {
4126 		rc = fxregs_fixup(&fx_state, size);
4127 		if (rc != X86EMUL_CONTINUE)
4128 			goto out;
4129 	}
4130 
4131 	if (fx_state.mxcsr >> 16) {
4132 		rc = emulate_gp(ctxt, 0);
4133 		goto out;
4134 	}
4135 
4136 	if (rc == X86EMUL_CONTINUE)
4137 		rc = asm_safe("fxrstor %[fx]", : [fx] "m"(fx_state));
4138 
4139 out:
4140 	kvm_fpu_put();
4141 
4142 	return rc;
4143 }
4144 
4145 static int em_xsetbv(struct x86_emulate_ctxt *ctxt)
4146 {
4147 	u32 eax, ecx, edx;
4148 
4149 	eax = reg_read(ctxt, VCPU_REGS_RAX);
4150 	edx = reg_read(ctxt, VCPU_REGS_RDX);
4151 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
4152 
4153 	if (ctxt->ops->set_xcr(ctxt, ecx, ((u64)edx << 32) | eax))
4154 		return emulate_gp(ctxt, 0);
4155 
4156 	return X86EMUL_CONTINUE;
4157 }
4158 
4159 static bool valid_cr(int nr)
4160 {
4161 	switch (nr) {
4162 	case 0:
4163 	case 2 ... 4:
4164 	case 8:
4165 		return true;
4166 	default:
4167 		return false;
4168 	}
4169 }
4170 
4171 static int check_cr_access(struct x86_emulate_ctxt *ctxt)
4172 {
4173 	if (!valid_cr(ctxt->modrm_reg))
4174 		return emulate_ud(ctxt);
4175 
4176 	return X86EMUL_CONTINUE;
4177 }
4178 
4179 static int check_dr7_gd(struct x86_emulate_ctxt *ctxt)
4180 {
4181 	unsigned long dr7;
4182 
4183 	ctxt->ops->get_dr(ctxt, 7, &dr7);
4184 
4185 	/* Check if DR7.Global_Enable is set */
4186 	return dr7 & (1 << 13);
4187 }
4188 
4189 static int check_dr_read(struct x86_emulate_ctxt *ctxt)
4190 {
4191 	int dr = ctxt->modrm_reg;
4192 	u64 cr4;
4193 
4194 	if (dr > 7)
4195 		return emulate_ud(ctxt);
4196 
4197 	cr4 = ctxt->ops->get_cr(ctxt, 4);
4198 	if ((cr4 & X86_CR4_DE) && (dr == 4 || dr == 5))
4199 		return emulate_ud(ctxt);
4200 
4201 	if (check_dr7_gd(ctxt)) {
4202 		ulong dr6;
4203 
4204 		ctxt->ops->get_dr(ctxt, 6, &dr6);
4205 		dr6 &= ~DR_TRAP_BITS;
4206 		dr6 |= DR6_BD | DR6_ACTIVE_LOW;
4207 		ctxt->ops->set_dr(ctxt, 6, dr6);
4208 		return emulate_db(ctxt);
4209 	}
4210 
4211 	return X86EMUL_CONTINUE;
4212 }
4213 
4214 static int check_dr_write(struct x86_emulate_ctxt *ctxt)
4215 {
4216 	u64 new_val = ctxt->src.val64;
4217 	int dr = ctxt->modrm_reg;
4218 
4219 	if ((dr == 6 || dr == 7) && (new_val & 0xffffffff00000000ULL))
4220 		return emulate_gp(ctxt, 0);
4221 
4222 	return check_dr_read(ctxt);
4223 }
4224 
4225 static int check_svme(struct x86_emulate_ctxt *ctxt)
4226 {
4227 	u64 efer = 0;
4228 
4229 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4230 
4231 	if (!(efer & EFER_SVME))
4232 		return emulate_ud(ctxt);
4233 
4234 	return X86EMUL_CONTINUE;
4235 }
4236 
4237 static int check_svme_pa(struct x86_emulate_ctxt *ctxt)
4238 {
4239 	u64 rax = reg_read(ctxt, VCPU_REGS_RAX);
4240 
4241 	/* Valid physical address? */
4242 	if (rax & 0xffff000000000000ULL)
4243 		return emulate_gp(ctxt, 0);
4244 
4245 	return check_svme(ctxt);
4246 }
4247 
4248 static int check_rdtsc(struct x86_emulate_ctxt *ctxt)
4249 {
4250 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4251 
4252 	if (cr4 & X86_CR4_TSD && ctxt->ops->cpl(ctxt))
4253 		return emulate_gp(ctxt, 0);
4254 
4255 	return X86EMUL_CONTINUE;
4256 }
4257 
4258 static int check_rdpmc(struct x86_emulate_ctxt *ctxt)
4259 {
4260 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4261 	u64 rcx = reg_read(ctxt, VCPU_REGS_RCX);
4262 
4263 	/*
4264 	 * VMware allows access to these Pseduo-PMCs even when read via RDPMC
4265 	 * in Ring3 when CR4.PCE=0.
4266 	 */
4267 	if (enable_vmware_backdoor && is_vmware_backdoor_pmc(rcx))
4268 		return X86EMUL_CONTINUE;
4269 
4270 	/*
4271 	 * If CR4.PCE is set, the SDM requires CPL=0 or CR0.PE=0.  The CR0.PE
4272 	 * check however is unnecessary because CPL is always 0 outside
4273 	 * protected mode.
4274 	 */
4275 	if ((!(cr4 & X86_CR4_PCE) && ctxt->ops->cpl(ctxt)) ||
4276 	    ctxt->ops->check_pmc(ctxt, rcx))
4277 		return emulate_gp(ctxt, 0);
4278 
4279 	return X86EMUL_CONTINUE;
4280 }
4281 
4282 static int check_perm_in(struct x86_emulate_ctxt *ctxt)
4283 {
4284 	ctxt->dst.bytes = min(ctxt->dst.bytes, 4u);
4285 	if (!emulator_io_permited(ctxt, ctxt->src.val, ctxt->dst.bytes))
4286 		return emulate_gp(ctxt, 0);
4287 
4288 	return X86EMUL_CONTINUE;
4289 }
4290 
4291 static int check_perm_out(struct x86_emulate_ctxt *ctxt)
4292 {
4293 	ctxt->src.bytes = min(ctxt->src.bytes, 4u);
4294 	if (!emulator_io_permited(ctxt, ctxt->dst.val, ctxt->src.bytes))
4295 		return emulate_gp(ctxt, 0);
4296 
4297 	return X86EMUL_CONTINUE;
4298 }
4299 
4300 #define D(_y) { .flags = (_y) }
4301 #define DI(_y, _i) { .flags = (_y)|Intercept, .intercept = x86_intercept_##_i }
4302 #define DIP(_y, _i, _p) { .flags = (_y)|Intercept|CheckPerm, \
4303 		      .intercept = x86_intercept_##_i, .check_perm = (_p) }
4304 #define N    D(NotImpl)
4305 #define EXT(_f, _e) { .flags = ((_f) | RMExt), .u.group = (_e) }
4306 #define G(_f, _g) { .flags = ((_f) | Group | ModRM), .u.group = (_g) }
4307 #define GD(_f, _g) { .flags = ((_f) | GroupDual | ModRM), .u.gdual = (_g) }
4308 #define ID(_f, _i) { .flags = ((_f) | InstrDual | ModRM), .u.idual = (_i) }
4309 #define MD(_f, _m) { .flags = ((_f) | ModeDual), .u.mdual = (_m) }
4310 #define E(_f, _e) { .flags = ((_f) | Escape | ModRM), .u.esc = (_e) }
4311 #define I(_f, _e) { .flags = (_f), .u.execute = (_e) }
4312 #define F(_f, _e) { .flags = (_f) | Fastop, .u.fastop = (_e) }
4313 #define II(_f, _e, _i) \
4314 	{ .flags = (_f)|Intercept, .u.execute = (_e), .intercept = x86_intercept_##_i }
4315 #define IIP(_f, _e, _i, _p) \
4316 	{ .flags = (_f)|Intercept|CheckPerm, .u.execute = (_e), \
4317 	  .intercept = x86_intercept_##_i, .check_perm = (_p) }
4318 #define GP(_f, _g) { .flags = ((_f) | Prefix), .u.gprefix = (_g) }
4319 
4320 #define D2bv(_f)      D((_f) | ByteOp), D(_f)
4321 #define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p)
4322 #define I2bv(_f, _e)  I((_f) | ByteOp, _e), I(_f, _e)
4323 #define F2bv(_f, _e)  F((_f) | ByteOp, _e), F(_f, _e)
4324 #define I2bvIP(_f, _e, _i, _p) \
4325 	IIP((_f) | ByteOp, _e, _i, _p), IIP(_f, _e, _i, _p)
4326 
4327 #define F6ALU(_f, _e) F2bv((_f) | DstMem | SrcReg | ModRM, _e),		\
4328 		F2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e),	\
4329 		F2bv(((_f) & ~Lock) | DstAcc | SrcImm, _e)
4330 
4331 static const struct opcode group7_rm0[] = {
4332 	N,
4333 	I(SrcNone | Priv | EmulateOnUD,	em_hypercall),
4334 	N, N, N, N, N, N,
4335 };
4336 
4337 static const struct opcode group7_rm1[] = {
4338 	DI(SrcNone | Priv, monitor),
4339 	DI(SrcNone | Priv, mwait),
4340 	N, N, N, N, N, N,
4341 };
4342 
4343 static const struct opcode group7_rm2[] = {
4344 	N,
4345 	II(ImplicitOps | Priv,			em_xsetbv,	xsetbv),
4346 	N, N, N, N, N, N,
4347 };
4348 
4349 static const struct opcode group7_rm3[] = {
4350 	DIP(SrcNone | Prot | Priv,		vmrun,		check_svme_pa),
4351 	II(SrcNone  | Prot | EmulateOnUD,	em_hypercall,	vmmcall),
4352 	DIP(SrcNone | Prot | Priv,		vmload,		check_svme_pa),
4353 	DIP(SrcNone | Prot | Priv,		vmsave,		check_svme_pa),
4354 	DIP(SrcNone | Prot | Priv,		stgi,		check_svme),
4355 	DIP(SrcNone | Prot | Priv,		clgi,		check_svme),
4356 	DIP(SrcNone | Prot | Priv,		skinit,		check_svme),
4357 	DIP(SrcNone | Prot | Priv,		invlpga,	check_svme),
4358 };
4359 
4360 static const struct opcode group7_rm7[] = {
4361 	N,
4362 	DIP(SrcNone, rdtscp, check_rdtsc),
4363 	N, N, N, N, N, N,
4364 };
4365 
4366 static const struct opcode group1[] = {
4367 	F(Lock, em_add),
4368 	F(Lock | PageTable, em_or),
4369 	F(Lock, em_adc),
4370 	F(Lock, em_sbb),
4371 	F(Lock | PageTable, em_and),
4372 	F(Lock, em_sub),
4373 	F(Lock, em_xor),
4374 	F(NoWrite, em_cmp),
4375 };
4376 
4377 static const struct opcode group1A[] = {
4378 	I(DstMem | SrcNone | Mov | Stack | IncSP | TwoMemOp, em_pop), N, N, N, N, N, N, N,
4379 };
4380 
4381 static const struct opcode group2[] = {
4382 	F(DstMem | ModRM, em_rol),
4383 	F(DstMem | ModRM, em_ror),
4384 	F(DstMem | ModRM, em_rcl),
4385 	F(DstMem | ModRM, em_rcr),
4386 	F(DstMem | ModRM, em_shl),
4387 	F(DstMem | ModRM, em_shr),
4388 	F(DstMem | ModRM, em_shl),
4389 	F(DstMem | ModRM, em_sar),
4390 };
4391 
4392 static const struct opcode group3[] = {
4393 	F(DstMem | SrcImm | NoWrite, em_test),
4394 	F(DstMem | SrcImm | NoWrite, em_test),
4395 	F(DstMem | SrcNone | Lock, em_not),
4396 	F(DstMem | SrcNone | Lock, em_neg),
4397 	F(DstXacc | Src2Mem, em_mul_ex),
4398 	F(DstXacc | Src2Mem, em_imul_ex),
4399 	F(DstXacc | Src2Mem, em_div_ex),
4400 	F(DstXacc | Src2Mem, em_idiv_ex),
4401 };
4402 
4403 static const struct opcode group4[] = {
4404 	F(ByteOp | DstMem | SrcNone | Lock, em_inc),
4405 	F(ByteOp | DstMem | SrcNone | Lock, em_dec),
4406 	N, N, N, N, N, N,
4407 };
4408 
4409 static const struct opcode group5[] = {
4410 	F(DstMem | SrcNone | Lock,		em_inc),
4411 	F(DstMem | SrcNone | Lock,		em_dec),
4412 	I(SrcMem | NearBranch | IsBranch,       em_call_near_abs),
4413 	I(SrcMemFAddr | ImplicitOps | IsBranch, em_call_far),
4414 	I(SrcMem | NearBranch | IsBranch,       em_jmp_abs),
4415 	I(SrcMemFAddr | ImplicitOps | IsBranch, em_jmp_far),
4416 	I(SrcMem | Stack | TwoMemOp,		em_push), D(Undefined),
4417 };
4418 
4419 static const struct opcode group6[] = {
4420 	II(Prot | DstMem,	   em_sldt, sldt),
4421 	II(Prot | DstMem,	   em_str, str),
4422 	II(Prot | Priv | SrcMem16, em_lldt, lldt),
4423 	II(Prot | Priv | SrcMem16, em_ltr, ltr),
4424 	N, N, N, N,
4425 };
4426 
4427 static const struct group_dual group7 = { {
4428 	II(Mov | DstMem,			em_sgdt, sgdt),
4429 	II(Mov | DstMem,			em_sidt, sidt),
4430 	II(SrcMem | Priv,			em_lgdt, lgdt),
4431 	II(SrcMem | Priv,			em_lidt, lidt),
4432 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4433 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4434 	II(SrcMem | ByteOp | Priv | NoAccess,	em_invlpg, invlpg),
4435 }, {
4436 	EXT(0, group7_rm0),
4437 	EXT(0, group7_rm1),
4438 	EXT(0, group7_rm2),
4439 	EXT(0, group7_rm3),
4440 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4441 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4442 	EXT(0, group7_rm7),
4443 } };
4444 
4445 static const struct opcode group8[] = {
4446 	N, N, N, N,
4447 	F(DstMem | SrcImmByte | NoWrite,		em_bt),
4448 	F(DstMem | SrcImmByte | Lock | PageTable,	em_bts),
4449 	F(DstMem | SrcImmByte | Lock,			em_btr),
4450 	F(DstMem | SrcImmByte | Lock | PageTable,	em_btc),
4451 };
4452 
4453 /*
4454  * The "memory" destination is actually always a register, since we come
4455  * from the register case of group9.
4456  */
4457 static const struct gprefix pfx_0f_c7_7 = {
4458 	N, N, N, II(DstMem | ModRM | Op3264 | EmulateOnUD, em_rdpid, rdpid),
4459 };
4460 
4461 
4462 static const struct group_dual group9 = { {
4463 	N, I(DstMem64 | Lock | PageTable, em_cmpxchg8b), N, N, N, N, N, N,
4464 }, {
4465 	N, N, N, N, N, N, N,
4466 	GP(0, &pfx_0f_c7_7),
4467 } };
4468 
4469 static const struct opcode group11[] = {
4470 	I(DstMem | SrcImm | Mov | PageTable, em_mov),
4471 	X7(D(Undefined)),
4472 };
4473 
4474 static const struct gprefix pfx_0f_ae_7 = {
4475 	I(SrcMem | ByteOp, em_clflush), I(SrcMem | ByteOp, em_clflushopt), N, N,
4476 };
4477 
4478 static const struct group_dual group15 = { {
4479 	I(ModRM | Aligned16, em_fxsave),
4480 	I(ModRM | Aligned16, em_fxrstor),
4481 	N, N, N, N, N, GP(0, &pfx_0f_ae_7),
4482 }, {
4483 	N, N, N, N, N, N, N, N,
4484 } };
4485 
4486 static const struct gprefix pfx_0f_6f_0f_7f = {
4487 	I(Mmx, em_mov), I(Sse | Aligned, em_mov), N, I(Sse | Unaligned, em_mov),
4488 };
4489 
4490 static const struct instr_dual instr_dual_0f_2b = {
4491 	I(0, em_mov), N
4492 };
4493 
4494 static const struct gprefix pfx_0f_2b = {
4495 	ID(0, &instr_dual_0f_2b), ID(0, &instr_dual_0f_2b), N, N,
4496 };
4497 
4498 static const struct gprefix pfx_0f_10_0f_11 = {
4499 	I(Unaligned, em_mov), I(Unaligned, em_mov), N, N,
4500 };
4501 
4502 static const struct gprefix pfx_0f_28_0f_29 = {
4503 	I(Aligned, em_mov), I(Aligned, em_mov), N, N,
4504 };
4505 
4506 static const struct gprefix pfx_0f_e7 = {
4507 	N, I(Sse, em_mov), N, N,
4508 };
4509 
4510 static const struct escape escape_d9 = { {
4511 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstcw),
4512 }, {
4513 	/* 0xC0 - 0xC7 */
4514 	N, N, N, N, N, N, N, N,
4515 	/* 0xC8 - 0xCF */
4516 	N, N, N, N, N, N, N, N,
4517 	/* 0xD0 - 0xC7 */
4518 	N, N, N, N, N, N, N, N,
4519 	/* 0xD8 - 0xDF */
4520 	N, N, N, N, N, N, N, N,
4521 	/* 0xE0 - 0xE7 */
4522 	N, N, N, N, N, N, N, N,
4523 	/* 0xE8 - 0xEF */
4524 	N, N, N, N, N, N, N, N,
4525 	/* 0xF0 - 0xF7 */
4526 	N, N, N, N, N, N, N, N,
4527 	/* 0xF8 - 0xFF */
4528 	N, N, N, N, N, N, N, N,
4529 } };
4530 
4531 static const struct escape escape_db = { {
4532 	N, N, N, N, N, N, N, N,
4533 }, {
4534 	/* 0xC0 - 0xC7 */
4535 	N, N, N, N, N, N, N, N,
4536 	/* 0xC8 - 0xCF */
4537 	N, N, N, N, N, N, N, N,
4538 	/* 0xD0 - 0xC7 */
4539 	N, N, N, N, N, N, N, N,
4540 	/* 0xD8 - 0xDF */
4541 	N, N, N, N, N, N, N, N,
4542 	/* 0xE0 - 0xE7 */
4543 	N, N, N, I(ImplicitOps, em_fninit), N, N, N, N,
4544 	/* 0xE8 - 0xEF */
4545 	N, N, N, N, N, N, N, N,
4546 	/* 0xF0 - 0xF7 */
4547 	N, N, N, N, N, N, N, N,
4548 	/* 0xF8 - 0xFF */
4549 	N, N, N, N, N, N, N, N,
4550 } };
4551 
4552 static const struct escape escape_dd = { {
4553 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstsw),
4554 }, {
4555 	/* 0xC0 - 0xC7 */
4556 	N, N, N, N, N, N, N, N,
4557 	/* 0xC8 - 0xCF */
4558 	N, N, N, N, N, N, N, N,
4559 	/* 0xD0 - 0xC7 */
4560 	N, N, N, N, N, N, N, N,
4561 	/* 0xD8 - 0xDF */
4562 	N, N, N, N, N, N, N, N,
4563 	/* 0xE0 - 0xE7 */
4564 	N, N, N, N, N, N, N, N,
4565 	/* 0xE8 - 0xEF */
4566 	N, N, N, N, N, N, N, N,
4567 	/* 0xF0 - 0xF7 */
4568 	N, N, N, N, N, N, N, N,
4569 	/* 0xF8 - 0xFF */
4570 	N, N, N, N, N, N, N, N,
4571 } };
4572 
4573 static const struct instr_dual instr_dual_0f_c3 = {
4574 	I(DstMem | SrcReg | ModRM | No16 | Mov, em_mov), N
4575 };
4576 
4577 static const struct mode_dual mode_dual_63 = {
4578 	N, I(DstReg | SrcMem32 | ModRM | Mov, em_movsxd)
4579 };
4580 
4581 static const struct instr_dual instr_dual_8d = {
4582 	D(DstReg | SrcMem | ModRM | NoAccess), N
4583 };
4584 
4585 static const struct opcode opcode_table[256] = {
4586 	/* 0x00 - 0x07 */
4587 	F6ALU(Lock, em_add),
4588 	I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg),
4589 	I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg),
4590 	/* 0x08 - 0x0F */
4591 	F6ALU(Lock | PageTable, em_or),
4592 	I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg),
4593 	N,
4594 	/* 0x10 - 0x17 */
4595 	F6ALU(Lock, em_adc),
4596 	I(ImplicitOps | Stack | No64 | Src2SS, em_push_sreg),
4597 	I(ImplicitOps | Stack | No64 | Src2SS, em_pop_sreg),
4598 	/* 0x18 - 0x1F */
4599 	F6ALU(Lock, em_sbb),
4600 	I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg),
4601 	I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg),
4602 	/* 0x20 - 0x27 */
4603 	F6ALU(Lock | PageTable, em_and), N, N,
4604 	/* 0x28 - 0x2F */
4605 	F6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das),
4606 	/* 0x30 - 0x37 */
4607 	F6ALU(Lock, em_xor), N, N,
4608 	/* 0x38 - 0x3F */
4609 	F6ALU(NoWrite, em_cmp), N, N,
4610 	/* 0x40 - 0x4F */
4611 	X8(F(DstReg, em_inc)), X8(F(DstReg, em_dec)),
4612 	/* 0x50 - 0x57 */
4613 	X8(I(SrcReg | Stack, em_push)),
4614 	/* 0x58 - 0x5F */
4615 	X8(I(DstReg | Stack, em_pop)),
4616 	/* 0x60 - 0x67 */
4617 	I(ImplicitOps | Stack | No64, em_pusha),
4618 	I(ImplicitOps | Stack | No64, em_popa),
4619 	N, MD(ModRM, &mode_dual_63),
4620 	N, N, N, N,
4621 	/* 0x68 - 0x6F */
4622 	I(SrcImm | Mov | Stack, em_push),
4623 	I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op),
4624 	I(SrcImmByte | Mov | Stack, em_push),
4625 	I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op),
4626 	I2bvIP(DstDI | SrcDX | Mov | String | Unaligned, em_in, ins, check_perm_in), /* insb, insw/insd */
4627 	I2bvIP(SrcSI | DstDX | String, em_out, outs, check_perm_out), /* outsb, outsw/outsd */
4628 	/* 0x70 - 0x7F */
4629 	X16(D(SrcImmByte | NearBranch | IsBranch)),
4630 	/* 0x80 - 0x87 */
4631 	G(ByteOp | DstMem | SrcImm, group1),
4632 	G(DstMem | SrcImm, group1),
4633 	G(ByteOp | DstMem | SrcImm | No64, group1),
4634 	G(DstMem | SrcImmByte, group1),
4635 	F2bv(DstMem | SrcReg | ModRM | NoWrite, em_test),
4636 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_xchg),
4637 	/* 0x88 - 0x8F */
4638 	I2bv(DstMem | SrcReg | ModRM | Mov | PageTable, em_mov),
4639 	I2bv(DstReg | SrcMem | ModRM | Mov, em_mov),
4640 	I(DstMem | SrcNone | ModRM | Mov | PageTable, em_mov_rm_sreg),
4641 	ID(0, &instr_dual_8d),
4642 	I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm),
4643 	G(0, group1A),
4644 	/* 0x90 - 0x97 */
4645 	DI(SrcAcc | DstReg, pause), X7(D(SrcAcc | DstReg)),
4646 	/* 0x98 - 0x9F */
4647 	D(DstAcc | SrcNone), I(ImplicitOps | SrcAcc, em_cwd),
4648 	I(SrcImmFAddr | No64 | IsBranch, em_call_far), N,
4649 	II(ImplicitOps | Stack, em_pushf, pushf),
4650 	II(ImplicitOps | Stack, em_popf, popf),
4651 	I(ImplicitOps, em_sahf), I(ImplicitOps, em_lahf),
4652 	/* 0xA0 - 0xA7 */
4653 	I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov),
4654 	I2bv(DstMem | SrcAcc | Mov | MemAbs | PageTable, em_mov),
4655 	I2bv(SrcSI | DstDI | Mov | String | TwoMemOp, em_mov),
4656 	F2bv(SrcSI | DstDI | String | NoWrite | TwoMemOp, em_cmp_r),
4657 	/* 0xA8 - 0xAF */
4658 	F2bv(DstAcc | SrcImm | NoWrite, em_test),
4659 	I2bv(SrcAcc | DstDI | Mov | String, em_mov),
4660 	I2bv(SrcSI | DstAcc | Mov | String, em_mov),
4661 	F2bv(SrcAcc | DstDI | String | NoWrite, em_cmp_r),
4662 	/* 0xB0 - 0xB7 */
4663 	X8(I(ByteOp | DstReg | SrcImm | Mov, em_mov)),
4664 	/* 0xB8 - 0xBF */
4665 	X8(I(DstReg | SrcImm64 | Mov, em_mov)),
4666 	/* 0xC0 - 0xC7 */
4667 	G(ByteOp | Src2ImmByte, group2), G(Src2ImmByte, group2),
4668 	I(ImplicitOps | NearBranch | SrcImmU16 | IsBranch, em_ret_near_imm),
4669 	I(ImplicitOps | NearBranch | IsBranch, em_ret),
4670 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2ES, em_lseg),
4671 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2DS, em_lseg),
4672 	G(ByteOp, group11), G(0, group11),
4673 	/* 0xC8 - 0xCF */
4674 	I(Stack | SrcImmU16 | Src2ImmByte | IsBranch, em_enter),
4675 	I(Stack | IsBranch, em_leave),
4676 	I(ImplicitOps | SrcImmU16 | IsBranch, em_ret_far_imm),
4677 	I(ImplicitOps | IsBranch, em_ret_far),
4678 	D(ImplicitOps | IsBranch), DI(SrcImmByte | IsBranch, intn),
4679 	D(ImplicitOps | No64 | IsBranch),
4680 	II(ImplicitOps | IsBranch, em_iret, iret),
4681 	/* 0xD0 - 0xD7 */
4682 	G(Src2One | ByteOp, group2), G(Src2One, group2),
4683 	G(Src2CL | ByteOp, group2), G(Src2CL, group2),
4684 	I(DstAcc | SrcImmUByte | No64, em_aam),
4685 	I(DstAcc | SrcImmUByte | No64, em_aad),
4686 	F(DstAcc | ByteOp | No64, em_salc),
4687 	I(DstAcc | SrcXLat | ByteOp, em_mov),
4688 	/* 0xD8 - 0xDF */
4689 	N, E(0, &escape_d9), N, E(0, &escape_db), N, E(0, &escape_dd), N, N,
4690 	/* 0xE0 - 0xE7 */
4691 	X3(I(SrcImmByte | NearBranch | IsBranch, em_loop)),
4692 	I(SrcImmByte | NearBranch | IsBranch, em_jcxz),
4693 	I2bvIP(SrcImmUByte | DstAcc, em_in,  in,  check_perm_in),
4694 	I2bvIP(SrcAcc | DstImmUByte, em_out, out, check_perm_out),
4695 	/* 0xE8 - 0xEF */
4696 	I(SrcImm | NearBranch | IsBranch, em_call),
4697 	D(SrcImm | ImplicitOps | NearBranch | IsBranch),
4698 	I(SrcImmFAddr | No64 | IsBranch, em_jmp_far),
4699 	D(SrcImmByte | ImplicitOps | NearBranch | IsBranch),
4700 	I2bvIP(SrcDX | DstAcc, em_in,  in,  check_perm_in),
4701 	I2bvIP(SrcAcc | DstDX, em_out, out, check_perm_out),
4702 	/* 0xF0 - 0xF7 */
4703 	N, DI(ImplicitOps, icebp), N, N,
4704 	DI(ImplicitOps | Priv, hlt), D(ImplicitOps),
4705 	G(ByteOp, group3), G(0, group3),
4706 	/* 0xF8 - 0xFF */
4707 	D(ImplicitOps), D(ImplicitOps),
4708 	I(ImplicitOps, em_cli), I(ImplicitOps, em_sti),
4709 	D(ImplicitOps), D(ImplicitOps), G(0, group4), G(0, group5),
4710 };
4711 
4712 static const struct opcode twobyte_table[256] = {
4713 	/* 0x00 - 0x0F */
4714 	G(0, group6), GD(0, &group7), N, N,
4715 	N, I(ImplicitOps | EmulateOnUD | IsBranch, em_syscall),
4716 	II(ImplicitOps | Priv, em_clts, clts), N,
4717 	DI(ImplicitOps | Priv, invd), DI(ImplicitOps | Priv, wbinvd), N, N,
4718 	N, D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4719 	/* 0x10 - 0x1F */
4720 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_10_0f_11),
4721 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_10_0f_11),
4722 	N, N, N, N, N, N,
4723 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 4 * prefetch + 4 * reserved NOP */
4724 	D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4725 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4726 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4727 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4728 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* NOP + 7 * reserved NOP */
4729 	/* 0x20 - 0x2F */
4730 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, cr_read, check_cr_access),
4731 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, dr_read, check_dr_read),
4732 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_cr_write, cr_write,
4733 						check_cr_access),
4734 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_dr_write, dr_write,
4735 						check_dr_write),
4736 	N, N, N, N,
4737 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_28_0f_29),
4738 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_28_0f_29),
4739 	N, GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_2b),
4740 	N, N, N, N,
4741 	/* 0x30 - 0x3F */
4742 	II(ImplicitOps | Priv, em_wrmsr, wrmsr),
4743 	IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc),
4744 	II(ImplicitOps | Priv, em_rdmsr, rdmsr),
4745 	IIP(ImplicitOps, em_rdpmc, rdpmc, check_rdpmc),
4746 	I(ImplicitOps | EmulateOnUD | IsBranch, em_sysenter),
4747 	I(ImplicitOps | Priv | EmulateOnUD | IsBranch, em_sysexit),
4748 	N, N,
4749 	N, N, N, N, N, N, N, N,
4750 	/* 0x40 - 0x4F */
4751 	X16(D(DstReg | SrcMem | ModRM)),
4752 	/* 0x50 - 0x5F */
4753 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4754 	/* 0x60 - 0x6F */
4755 	N, N, N, N,
4756 	N, N, N, N,
4757 	N, N, N, N,
4758 	N, N, N, GP(SrcMem | DstReg | ModRM | Mov, &pfx_0f_6f_0f_7f),
4759 	/* 0x70 - 0x7F */
4760 	N, N, N, N,
4761 	N, N, N, N,
4762 	N, N, N, N,
4763 	N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_6f_0f_7f),
4764 	/* 0x80 - 0x8F */
4765 	X16(D(SrcImm | NearBranch | IsBranch)),
4766 	/* 0x90 - 0x9F */
4767 	X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)),
4768 	/* 0xA0 - 0xA7 */
4769 	I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg),
4770 	II(ImplicitOps, em_cpuid, cpuid),
4771 	F(DstMem | SrcReg | ModRM | BitOp | NoWrite, em_bt),
4772 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld),
4773 	F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
4774 	/* 0xA8 - 0xAF */
4775 	I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
4776 	II(EmulateOnUD | ImplicitOps, em_rsm, rsm),
4777 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
4778 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
4779 	F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
4780 	GD(0, &group15), F(DstReg | SrcMem | ModRM, em_imul),
4781 	/* 0xB0 - 0xB7 */
4782 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg),
4783 	I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg),
4784 	F(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr),
4785 	I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg),
4786 	I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg),
4787 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4788 	/* 0xB8 - 0xBF */
4789 	N, N,
4790 	G(BitOp, group8),
4791 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_btc),
4792 	I(DstReg | SrcMem | ModRM, em_bsf_c),
4793 	I(DstReg | SrcMem | ModRM, em_bsr_c),
4794 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4795 	/* 0xC0 - 0xC7 */
4796 	F2bv(DstMem | SrcReg | ModRM | SrcWrite | Lock, em_xadd),
4797 	N, ID(0, &instr_dual_0f_c3),
4798 	N, N, N, GD(0, &group9),
4799 	/* 0xC8 - 0xCF */
4800 	X8(I(DstReg, em_bswap)),
4801 	/* 0xD0 - 0xDF */
4802 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4803 	/* 0xE0 - 0xEF */
4804 	N, N, N, N, N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_e7),
4805 	N, N, N, N, N, N, N, N,
4806 	/* 0xF0 - 0xFF */
4807 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N
4808 };
4809 
4810 static const struct instr_dual instr_dual_0f_38_f0 = {
4811 	I(DstReg | SrcMem | Mov, em_movbe), N
4812 };
4813 
4814 static const struct instr_dual instr_dual_0f_38_f1 = {
4815 	I(DstMem | SrcReg | Mov, em_movbe), N
4816 };
4817 
4818 static const struct gprefix three_byte_0f_38_f0 = {
4819 	ID(0, &instr_dual_0f_38_f0), N, N, N
4820 };
4821 
4822 static const struct gprefix three_byte_0f_38_f1 = {
4823 	ID(0, &instr_dual_0f_38_f1), N, N, N
4824 };
4825 
4826 /*
4827  * Insns below are selected by the prefix which indexed by the third opcode
4828  * byte.
4829  */
4830 static const struct opcode opcode_map_0f_38[256] = {
4831 	/* 0x00 - 0x7f */
4832 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4833 	/* 0x80 - 0xef */
4834 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4835 	/* 0xf0 - 0xf1 */
4836 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f0),
4837 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f1),
4838 	/* 0xf2 - 0xff */
4839 	N, N, X4(N), X8(N)
4840 };
4841 
4842 #undef D
4843 #undef N
4844 #undef G
4845 #undef GD
4846 #undef I
4847 #undef GP
4848 #undef EXT
4849 #undef MD
4850 #undef ID
4851 
4852 #undef D2bv
4853 #undef D2bvIP
4854 #undef I2bv
4855 #undef I2bvIP
4856 #undef I6ALU
4857 
4858 static unsigned imm_size(struct x86_emulate_ctxt *ctxt)
4859 {
4860 	unsigned size;
4861 
4862 	size = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4863 	if (size == 8)
4864 		size = 4;
4865 	return size;
4866 }
4867 
4868 static int decode_imm(struct x86_emulate_ctxt *ctxt, struct operand *op,
4869 		      unsigned size, bool sign_extension)
4870 {
4871 	int rc = X86EMUL_CONTINUE;
4872 
4873 	op->type = OP_IMM;
4874 	op->bytes = size;
4875 	op->addr.mem.ea = ctxt->_eip;
4876 	/* NB. Immediates are sign-extended as necessary. */
4877 	switch (op->bytes) {
4878 	case 1:
4879 		op->val = insn_fetch(s8, ctxt);
4880 		break;
4881 	case 2:
4882 		op->val = insn_fetch(s16, ctxt);
4883 		break;
4884 	case 4:
4885 		op->val = insn_fetch(s32, ctxt);
4886 		break;
4887 	case 8:
4888 		op->val = insn_fetch(s64, ctxt);
4889 		break;
4890 	}
4891 	if (!sign_extension) {
4892 		switch (op->bytes) {
4893 		case 1:
4894 			op->val &= 0xff;
4895 			break;
4896 		case 2:
4897 			op->val &= 0xffff;
4898 			break;
4899 		case 4:
4900 			op->val &= 0xffffffff;
4901 			break;
4902 		}
4903 	}
4904 done:
4905 	return rc;
4906 }
4907 
4908 static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op,
4909 			  unsigned d)
4910 {
4911 	int rc = X86EMUL_CONTINUE;
4912 
4913 	switch (d) {
4914 	case OpReg:
4915 		decode_register_operand(ctxt, op);
4916 		break;
4917 	case OpImmUByte:
4918 		rc = decode_imm(ctxt, op, 1, false);
4919 		break;
4920 	case OpMem:
4921 		ctxt->memop.bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4922 	mem_common:
4923 		*op = ctxt->memop;
4924 		ctxt->memopp = op;
4925 		if (ctxt->d & BitOp)
4926 			fetch_bit_operand(ctxt);
4927 		op->orig_val = op->val;
4928 		break;
4929 	case OpMem64:
4930 		ctxt->memop.bytes = (ctxt->op_bytes == 8) ? 16 : 8;
4931 		goto mem_common;
4932 	case OpAcc:
4933 		op->type = OP_REG;
4934 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4935 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4936 		fetch_register_operand(op);
4937 		op->orig_val = op->val;
4938 		break;
4939 	case OpAccLo:
4940 		op->type = OP_REG;
4941 		op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes;
4942 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4943 		fetch_register_operand(op);
4944 		op->orig_val = op->val;
4945 		break;
4946 	case OpAccHi:
4947 		if (ctxt->d & ByteOp) {
4948 			op->type = OP_NONE;
4949 			break;
4950 		}
4951 		op->type = OP_REG;
4952 		op->bytes = ctxt->op_bytes;
4953 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4954 		fetch_register_operand(op);
4955 		op->orig_val = op->val;
4956 		break;
4957 	case OpDI:
4958 		op->type = OP_MEM;
4959 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4960 		op->addr.mem.ea =
4961 			register_address(ctxt, VCPU_REGS_RDI);
4962 		op->addr.mem.seg = VCPU_SREG_ES;
4963 		op->val = 0;
4964 		op->count = 1;
4965 		break;
4966 	case OpDX:
4967 		op->type = OP_REG;
4968 		op->bytes = 2;
4969 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4970 		fetch_register_operand(op);
4971 		break;
4972 	case OpCL:
4973 		op->type = OP_IMM;
4974 		op->bytes = 1;
4975 		op->val = reg_read(ctxt, VCPU_REGS_RCX) & 0xff;
4976 		break;
4977 	case OpImmByte:
4978 		rc = decode_imm(ctxt, op, 1, true);
4979 		break;
4980 	case OpOne:
4981 		op->type = OP_IMM;
4982 		op->bytes = 1;
4983 		op->val = 1;
4984 		break;
4985 	case OpImm:
4986 		rc = decode_imm(ctxt, op, imm_size(ctxt), true);
4987 		break;
4988 	case OpImm64:
4989 		rc = decode_imm(ctxt, op, ctxt->op_bytes, true);
4990 		break;
4991 	case OpMem8:
4992 		ctxt->memop.bytes = 1;
4993 		if (ctxt->memop.type == OP_REG) {
4994 			ctxt->memop.addr.reg = decode_register(ctxt,
4995 					ctxt->modrm_rm, true);
4996 			fetch_register_operand(&ctxt->memop);
4997 		}
4998 		goto mem_common;
4999 	case OpMem16:
5000 		ctxt->memop.bytes = 2;
5001 		goto mem_common;
5002 	case OpMem32:
5003 		ctxt->memop.bytes = 4;
5004 		goto mem_common;
5005 	case OpImmU16:
5006 		rc = decode_imm(ctxt, op, 2, false);
5007 		break;
5008 	case OpImmU:
5009 		rc = decode_imm(ctxt, op, imm_size(ctxt), false);
5010 		break;
5011 	case OpSI:
5012 		op->type = OP_MEM;
5013 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
5014 		op->addr.mem.ea =
5015 			register_address(ctxt, VCPU_REGS_RSI);
5016 		op->addr.mem.seg = ctxt->seg_override;
5017 		op->val = 0;
5018 		op->count = 1;
5019 		break;
5020 	case OpXLat:
5021 		op->type = OP_MEM;
5022 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
5023 		op->addr.mem.ea =
5024 			address_mask(ctxt,
5025 				reg_read(ctxt, VCPU_REGS_RBX) +
5026 				(reg_read(ctxt, VCPU_REGS_RAX) & 0xff));
5027 		op->addr.mem.seg = ctxt->seg_override;
5028 		op->val = 0;
5029 		break;
5030 	case OpImmFAddr:
5031 		op->type = OP_IMM;
5032 		op->addr.mem.ea = ctxt->_eip;
5033 		op->bytes = ctxt->op_bytes + 2;
5034 		insn_fetch_arr(op->valptr, op->bytes, ctxt);
5035 		break;
5036 	case OpMemFAddr:
5037 		ctxt->memop.bytes = ctxt->op_bytes + 2;
5038 		goto mem_common;
5039 	case OpES:
5040 		op->type = OP_IMM;
5041 		op->val = VCPU_SREG_ES;
5042 		break;
5043 	case OpCS:
5044 		op->type = OP_IMM;
5045 		op->val = VCPU_SREG_CS;
5046 		break;
5047 	case OpSS:
5048 		op->type = OP_IMM;
5049 		op->val = VCPU_SREG_SS;
5050 		break;
5051 	case OpDS:
5052 		op->type = OP_IMM;
5053 		op->val = VCPU_SREG_DS;
5054 		break;
5055 	case OpFS:
5056 		op->type = OP_IMM;
5057 		op->val = VCPU_SREG_FS;
5058 		break;
5059 	case OpGS:
5060 		op->type = OP_IMM;
5061 		op->val = VCPU_SREG_GS;
5062 		break;
5063 	case OpImplicit:
5064 		/* Special instructions do their own operand decoding. */
5065 	default:
5066 		op->type = OP_NONE; /* Disable writeback. */
5067 		break;
5068 	}
5069 
5070 done:
5071 	return rc;
5072 }
5073 
5074 int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len, int emulation_type)
5075 {
5076 	int rc = X86EMUL_CONTINUE;
5077 	int mode = ctxt->mode;
5078 	int def_op_bytes, def_ad_bytes, goffset, simd_prefix;
5079 	bool op_prefix = false;
5080 	bool has_seg_override = false;
5081 	struct opcode opcode;
5082 	u16 dummy;
5083 	struct desc_struct desc;
5084 
5085 	ctxt->memop.type = OP_NONE;
5086 	ctxt->memopp = NULL;
5087 	ctxt->_eip = ctxt->eip;
5088 	ctxt->fetch.ptr = ctxt->fetch.data;
5089 	ctxt->fetch.end = ctxt->fetch.data + insn_len;
5090 	ctxt->opcode_len = 1;
5091 	ctxt->intercept = x86_intercept_none;
5092 	if (insn_len > 0)
5093 		memcpy(ctxt->fetch.data, insn, insn_len);
5094 	else {
5095 		rc = __do_insn_fetch_bytes(ctxt, 1);
5096 		if (rc != X86EMUL_CONTINUE)
5097 			goto done;
5098 	}
5099 
5100 	switch (mode) {
5101 	case X86EMUL_MODE_REAL:
5102 	case X86EMUL_MODE_VM86:
5103 		def_op_bytes = def_ad_bytes = 2;
5104 		ctxt->ops->get_segment(ctxt, &dummy, &desc, NULL, VCPU_SREG_CS);
5105 		if (desc.d)
5106 			def_op_bytes = def_ad_bytes = 4;
5107 		break;
5108 	case X86EMUL_MODE_PROT16:
5109 		def_op_bytes = def_ad_bytes = 2;
5110 		break;
5111 	case X86EMUL_MODE_PROT32:
5112 		def_op_bytes = def_ad_bytes = 4;
5113 		break;
5114 #ifdef CONFIG_X86_64
5115 	case X86EMUL_MODE_PROT64:
5116 		def_op_bytes = 4;
5117 		def_ad_bytes = 8;
5118 		break;
5119 #endif
5120 	default:
5121 		return EMULATION_FAILED;
5122 	}
5123 
5124 	ctxt->op_bytes = def_op_bytes;
5125 	ctxt->ad_bytes = def_ad_bytes;
5126 
5127 	/* Legacy prefixes. */
5128 	for (;;) {
5129 		switch (ctxt->b = insn_fetch(u8, ctxt)) {
5130 		case 0x66:	/* operand-size override */
5131 			op_prefix = true;
5132 			/* switch between 2/4 bytes */
5133 			ctxt->op_bytes = def_op_bytes ^ 6;
5134 			break;
5135 		case 0x67:	/* address-size override */
5136 			if (mode == X86EMUL_MODE_PROT64)
5137 				/* switch between 4/8 bytes */
5138 				ctxt->ad_bytes = def_ad_bytes ^ 12;
5139 			else
5140 				/* switch between 2/4 bytes */
5141 				ctxt->ad_bytes = def_ad_bytes ^ 6;
5142 			break;
5143 		case 0x26:	/* ES override */
5144 			has_seg_override = true;
5145 			ctxt->seg_override = VCPU_SREG_ES;
5146 			break;
5147 		case 0x2e:	/* CS override */
5148 			has_seg_override = true;
5149 			ctxt->seg_override = VCPU_SREG_CS;
5150 			break;
5151 		case 0x36:	/* SS override */
5152 			has_seg_override = true;
5153 			ctxt->seg_override = VCPU_SREG_SS;
5154 			break;
5155 		case 0x3e:	/* DS override */
5156 			has_seg_override = true;
5157 			ctxt->seg_override = VCPU_SREG_DS;
5158 			break;
5159 		case 0x64:	/* FS override */
5160 			has_seg_override = true;
5161 			ctxt->seg_override = VCPU_SREG_FS;
5162 			break;
5163 		case 0x65:	/* GS override */
5164 			has_seg_override = true;
5165 			ctxt->seg_override = VCPU_SREG_GS;
5166 			break;
5167 		case 0x40 ... 0x4f: /* REX */
5168 			if (mode != X86EMUL_MODE_PROT64)
5169 				goto done_prefixes;
5170 			ctxt->rex_prefix = ctxt->b;
5171 			continue;
5172 		case 0xf0:	/* LOCK */
5173 			ctxt->lock_prefix = 1;
5174 			break;
5175 		case 0xf2:	/* REPNE/REPNZ */
5176 		case 0xf3:	/* REP/REPE/REPZ */
5177 			ctxt->rep_prefix = ctxt->b;
5178 			break;
5179 		default:
5180 			goto done_prefixes;
5181 		}
5182 
5183 		/* Any legacy prefix after a REX prefix nullifies its effect. */
5184 
5185 		ctxt->rex_prefix = 0;
5186 	}
5187 
5188 done_prefixes:
5189 
5190 	/* REX prefix. */
5191 	if (ctxt->rex_prefix & 8)
5192 		ctxt->op_bytes = 8;	/* REX.W */
5193 
5194 	/* Opcode byte(s). */
5195 	opcode = opcode_table[ctxt->b];
5196 	/* Two-byte opcode? */
5197 	if (ctxt->b == 0x0f) {
5198 		ctxt->opcode_len = 2;
5199 		ctxt->b = insn_fetch(u8, ctxt);
5200 		opcode = twobyte_table[ctxt->b];
5201 
5202 		/* 0F_38 opcode map */
5203 		if (ctxt->b == 0x38) {
5204 			ctxt->opcode_len = 3;
5205 			ctxt->b = insn_fetch(u8, ctxt);
5206 			opcode = opcode_map_0f_38[ctxt->b];
5207 		}
5208 	}
5209 	ctxt->d = opcode.flags;
5210 
5211 	if (ctxt->d & ModRM)
5212 		ctxt->modrm = insn_fetch(u8, ctxt);
5213 
5214 	/* vex-prefix instructions are not implemented */
5215 	if (ctxt->opcode_len == 1 && (ctxt->b == 0xc5 || ctxt->b == 0xc4) &&
5216 	    (mode == X86EMUL_MODE_PROT64 || (ctxt->modrm & 0xc0) == 0xc0)) {
5217 		ctxt->d = NotImpl;
5218 	}
5219 
5220 	while (ctxt->d & GroupMask) {
5221 		switch (ctxt->d & GroupMask) {
5222 		case Group:
5223 			goffset = (ctxt->modrm >> 3) & 7;
5224 			opcode = opcode.u.group[goffset];
5225 			break;
5226 		case GroupDual:
5227 			goffset = (ctxt->modrm >> 3) & 7;
5228 			if ((ctxt->modrm >> 6) == 3)
5229 				opcode = opcode.u.gdual->mod3[goffset];
5230 			else
5231 				opcode = opcode.u.gdual->mod012[goffset];
5232 			break;
5233 		case RMExt:
5234 			goffset = ctxt->modrm & 7;
5235 			opcode = opcode.u.group[goffset];
5236 			break;
5237 		case Prefix:
5238 			if (ctxt->rep_prefix && op_prefix)
5239 				return EMULATION_FAILED;
5240 			simd_prefix = op_prefix ? 0x66 : ctxt->rep_prefix;
5241 			switch (simd_prefix) {
5242 			case 0x00: opcode = opcode.u.gprefix->pfx_no; break;
5243 			case 0x66: opcode = opcode.u.gprefix->pfx_66; break;
5244 			case 0xf2: opcode = opcode.u.gprefix->pfx_f2; break;
5245 			case 0xf3: opcode = opcode.u.gprefix->pfx_f3; break;
5246 			}
5247 			break;
5248 		case Escape:
5249 			if (ctxt->modrm > 0xbf) {
5250 				size_t size = ARRAY_SIZE(opcode.u.esc->high);
5251 				u32 index = array_index_nospec(
5252 					ctxt->modrm - 0xc0, size);
5253 
5254 				opcode = opcode.u.esc->high[index];
5255 			} else {
5256 				opcode = opcode.u.esc->op[(ctxt->modrm >> 3) & 7];
5257 			}
5258 			break;
5259 		case InstrDual:
5260 			if ((ctxt->modrm >> 6) == 3)
5261 				opcode = opcode.u.idual->mod3;
5262 			else
5263 				opcode = opcode.u.idual->mod012;
5264 			break;
5265 		case ModeDual:
5266 			if (ctxt->mode == X86EMUL_MODE_PROT64)
5267 				opcode = opcode.u.mdual->mode64;
5268 			else
5269 				opcode = opcode.u.mdual->mode32;
5270 			break;
5271 		default:
5272 			return EMULATION_FAILED;
5273 		}
5274 
5275 		ctxt->d &= ~(u64)GroupMask;
5276 		ctxt->d |= opcode.flags;
5277 	}
5278 
5279 	ctxt->is_branch = opcode.flags & IsBranch;
5280 
5281 	/* Unrecognised? */
5282 	if (ctxt->d == 0)
5283 		return EMULATION_FAILED;
5284 
5285 	ctxt->execute = opcode.u.execute;
5286 
5287 	if (unlikely(emulation_type & EMULTYPE_TRAP_UD) &&
5288 	    likely(!(ctxt->d & EmulateOnUD)))
5289 		return EMULATION_FAILED;
5290 
5291 	if (unlikely(ctxt->d &
5292 	    (NotImpl|Stack|Op3264|Sse|Mmx|Intercept|CheckPerm|NearBranch|
5293 	     No16))) {
5294 		/*
5295 		 * These are copied unconditionally here, and checked unconditionally
5296 		 * in x86_emulate_insn.
5297 		 */
5298 		ctxt->check_perm = opcode.check_perm;
5299 		ctxt->intercept = opcode.intercept;
5300 
5301 		if (ctxt->d & NotImpl)
5302 			return EMULATION_FAILED;
5303 
5304 		if (mode == X86EMUL_MODE_PROT64) {
5305 			if (ctxt->op_bytes == 4 && (ctxt->d & Stack))
5306 				ctxt->op_bytes = 8;
5307 			else if (ctxt->d & NearBranch)
5308 				ctxt->op_bytes = 8;
5309 		}
5310 
5311 		if (ctxt->d & Op3264) {
5312 			if (mode == X86EMUL_MODE_PROT64)
5313 				ctxt->op_bytes = 8;
5314 			else
5315 				ctxt->op_bytes = 4;
5316 		}
5317 
5318 		if ((ctxt->d & No16) && ctxt->op_bytes == 2)
5319 			ctxt->op_bytes = 4;
5320 
5321 		if (ctxt->d & Sse)
5322 			ctxt->op_bytes = 16;
5323 		else if (ctxt->d & Mmx)
5324 			ctxt->op_bytes = 8;
5325 	}
5326 
5327 	/* ModRM and SIB bytes. */
5328 	if (ctxt->d & ModRM) {
5329 		rc = decode_modrm(ctxt, &ctxt->memop);
5330 		if (!has_seg_override) {
5331 			has_seg_override = true;
5332 			ctxt->seg_override = ctxt->modrm_seg;
5333 		}
5334 	} else if (ctxt->d & MemAbs)
5335 		rc = decode_abs(ctxt, &ctxt->memop);
5336 	if (rc != X86EMUL_CONTINUE)
5337 		goto done;
5338 
5339 	if (!has_seg_override)
5340 		ctxt->seg_override = VCPU_SREG_DS;
5341 
5342 	ctxt->memop.addr.mem.seg = ctxt->seg_override;
5343 
5344 	/*
5345 	 * Decode and fetch the source operand: register, memory
5346 	 * or immediate.
5347 	 */
5348 	rc = decode_operand(ctxt, &ctxt->src, (ctxt->d >> SrcShift) & OpMask);
5349 	if (rc != X86EMUL_CONTINUE)
5350 		goto done;
5351 
5352 	/*
5353 	 * Decode and fetch the second source operand: register, memory
5354 	 * or immediate.
5355 	 */
5356 	rc = decode_operand(ctxt, &ctxt->src2, (ctxt->d >> Src2Shift) & OpMask);
5357 	if (rc != X86EMUL_CONTINUE)
5358 		goto done;
5359 
5360 	/* Decode and fetch the destination operand: register or memory. */
5361 	rc = decode_operand(ctxt, &ctxt->dst, (ctxt->d >> DstShift) & OpMask);
5362 
5363 	if (ctxt->rip_relative && likely(ctxt->memopp))
5364 		ctxt->memopp->addr.mem.ea = address_mask(ctxt,
5365 					ctxt->memopp->addr.mem.ea + ctxt->_eip);
5366 
5367 done:
5368 	if (rc == X86EMUL_PROPAGATE_FAULT)
5369 		ctxt->have_exception = true;
5370 	return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK;
5371 }
5372 
5373 bool x86_page_table_writing_insn(struct x86_emulate_ctxt *ctxt)
5374 {
5375 	return ctxt->d & PageTable;
5376 }
5377 
5378 static bool string_insn_completed(struct x86_emulate_ctxt *ctxt)
5379 {
5380 	/* The second termination condition only applies for REPE
5381 	 * and REPNE. Test if the repeat string operation prefix is
5382 	 * REPE/REPZ or REPNE/REPNZ and if it's the case it tests the
5383 	 * corresponding termination condition according to:
5384 	 * 	- if REPE/REPZ and ZF = 0 then done
5385 	 * 	- if REPNE/REPNZ and ZF = 1 then done
5386 	 */
5387 	if (((ctxt->b == 0xa6) || (ctxt->b == 0xa7) ||
5388 	     (ctxt->b == 0xae) || (ctxt->b == 0xaf))
5389 	    && (((ctxt->rep_prefix == REPE_PREFIX) &&
5390 		 ((ctxt->eflags & X86_EFLAGS_ZF) == 0))
5391 		|| ((ctxt->rep_prefix == REPNE_PREFIX) &&
5392 		    ((ctxt->eflags & X86_EFLAGS_ZF) == X86_EFLAGS_ZF))))
5393 		return true;
5394 
5395 	return false;
5396 }
5397 
5398 static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
5399 {
5400 	int rc;
5401 
5402 	kvm_fpu_get();
5403 	rc = asm_safe("fwait");
5404 	kvm_fpu_put();
5405 
5406 	if (unlikely(rc != X86EMUL_CONTINUE))
5407 		return emulate_exception(ctxt, MF_VECTOR, 0, false);
5408 
5409 	return X86EMUL_CONTINUE;
5410 }
5411 
5412 static void fetch_possible_mmx_operand(struct operand *op)
5413 {
5414 	if (op->type == OP_MM)
5415 		kvm_read_mmx_reg(op->addr.mm, &op->mm_val);
5416 }
5417 
5418 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop)
5419 {
5420 	ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF;
5421 
5422 	if (!(ctxt->d & ByteOp))
5423 		fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE;
5424 
5425 	asm("push %[flags]; popf; " CALL_NOSPEC " ; pushf; pop %[flags]\n"
5426 	    : "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags),
5427 	      [thunk_target]"+S"(fop), ASM_CALL_CONSTRAINT
5428 	    : "c"(ctxt->src2.val));
5429 
5430 	ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK);
5431 	if (!fop) /* exception is returned in fop variable */
5432 		return emulate_de(ctxt);
5433 	return X86EMUL_CONTINUE;
5434 }
5435 
5436 void init_decode_cache(struct x86_emulate_ctxt *ctxt)
5437 {
5438 	/* Clear fields that are set conditionally but read without a guard. */
5439 	ctxt->rip_relative = false;
5440 	ctxt->rex_prefix = 0;
5441 	ctxt->lock_prefix = 0;
5442 	ctxt->rep_prefix = 0;
5443 	ctxt->regs_valid = 0;
5444 	ctxt->regs_dirty = 0;
5445 
5446 	ctxt->io_read.pos = 0;
5447 	ctxt->io_read.end = 0;
5448 	ctxt->mem_read.end = 0;
5449 }
5450 
5451 int x86_emulate_insn(struct x86_emulate_ctxt *ctxt)
5452 {
5453 	const struct x86_emulate_ops *ops = ctxt->ops;
5454 	int rc = X86EMUL_CONTINUE;
5455 	int saved_dst_type = ctxt->dst.type;
5456 	unsigned emul_flags;
5457 
5458 	ctxt->mem_read.pos = 0;
5459 
5460 	/* LOCK prefix is allowed only with some instructions */
5461 	if (ctxt->lock_prefix && (!(ctxt->d & Lock) || ctxt->dst.type != OP_MEM)) {
5462 		rc = emulate_ud(ctxt);
5463 		goto done;
5464 	}
5465 
5466 	if ((ctxt->d & SrcMask) == SrcMemFAddr && ctxt->src.type != OP_MEM) {
5467 		rc = emulate_ud(ctxt);
5468 		goto done;
5469 	}
5470 
5471 	emul_flags = ctxt->ops->get_hflags(ctxt);
5472 	if (unlikely(ctxt->d &
5473 		     (No64|Undefined|Sse|Mmx|Intercept|CheckPerm|Priv|Prot|String))) {
5474 		if ((ctxt->mode == X86EMUL_MODE_PROT64 && (ctxt->d & No64)) ||
5475 				(ctxt->d & Undefined)) {
5476 			rc = emulate_ud(ctxt);
5477 			goto done;
5478 		}
5479 
5480 		if (((ctxt->d & (Sse|Mmx)) && ((ops->get_cr(ctxt, 0) & X86_CR0_EM)))
5481 		    || ((ctxt->d & Sse) && !(ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR))) {
5482 			rc = emulate_ud(ctxt);
5483 			goto done;
5484 		}
5485 
5486 		if ((ctxt->d & (Sse|Mmx)) && (ops->get_cr(ctxt, 0) & X86_CR0_TS)) {
5487 			rc = emulate_nm(ctxt);
5488 			goto done;
5489 		}
5490 
5491 		if (ctxt->d & Mmx) {
5492 			rc = flush_pending_x87_faults(ctxt);
5493 			if (rc != X86EMUL_CONTINUE)
5494 				goto done;
5495 			/*
5496 			 * Now that we know the fpu is exception safe, we can fetch
5497 			 * operands from it.
5498 			 */
5499 			fetch_possible_mmx_operand(&ctxt->src);
5500 			fetch_possible_mmx_operand(&ctxt->src2);
5501 			if (!(ctxt->d & Mov))
5502 				fetch_possible_mmx_operand(&ctxt->dst);
5503 		}
5504 
5505 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && ctxt->intercept) {
5506 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5507 						      X86_ICPT_PRE_EXCEPT);
5508 			if (rc != X86EMUL_CONTINUE)
5509 				goto done;
5510 		}
5511 
5512 		/* Instruction can only be executed in protected mode */
5513 		if ((ctxt->d & Prot) && ctxt->mode < X86EMUL_MODE_PROT16) {
5514 			rc = emulate_ud(ctxt);
5515 			goto done;
5516 		}
5517 
5518 		/* Privileged instruction can be executed only in CPL=0 */
5519 		if ((ctxt->d & Priv) && ops->cpl(ctxt)) {
5520 			if (ctxt->d & PrivUD)
5521 				rc = emulate_ud(ctxt);
5522 			else
5523 				rc = emulate_gp(ctxt, 0);
5524 			goto done;
5525 		}
5526 
5527 		/* Do instruction specific permission checks */
5528 		if (ctxt->d & CheckPerm) {
5529 			rc = ctxt->check_perm(ctxt);
5530 			if (rc != X86EMUL_CONTINUE)
5531 				goto done;
5532 		}
5533 
5534 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5535 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5536 						      X86_ICPT_POST_EXCEPT);
5537 			if (rc != X86EMUL_CONTINUE)
5538 				goto done;
5539 		}
5540 
5541 		if (ctxt->rep_prefix && (ctxt->d & String)) {
5542 			/* All REP prefixes have the same first termination condition */
5543 			if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0) {
5544 				string_registers_quirk(ctxt);
5545 				ctxt->eip = ctxt->_eip;
5546 				ctxt->eflags &= ~X86_EFLAGS_RF;
5547 				goto done;
5548 			}
5549 		}
5550 	}
5551 
5552 	if ((ctxt->src.type == OP_MEM) && !(ctxt->d & NoAccess)) {
5553 		rc = segmented_read(ctxt, ctxt->src.addr.mem,
5554 				    ctxt->src.valptr, ctxt->src.bytes);
5555 		if (rc != X86EMUL_CONTINUE)
5556 			goto done;
5557 		ctxt->src.orig_val64 = ctxt->src.val64;
5558 	}
5559 
5560 	if (ctxt->src2.type == OP_MEM) {
5561 		rc = segmented_read(ctxt, ctxt->src2.addr.mem,
5562 				    &ctxt->src2.val, ctxt->src2.bytes);
5563 		if (rc != X86EMUL_CONTINUE)
5564 			goto done;
5565 	}
5566 
5567 	if ((ctxt->d & DstMask) == ImplicitOps)
5568 		goto special_insn;
5569 
5570 
5571 	if ((ctxt->dst.type == OP_MEM) && !(ctxt->d & Mov)) {
5572 		/* optimisation - avoid slow emulated read if Mov */
5573 		rc = segmented_read(ctxt, ctxt->dst.addr.mem,
5574 				   &ctxt->dst.val, ctxt->dst.bytes);
5575 		if (rc != X86EMUL_CONTINUE) {
5576 			if (!(ctxt->d & NoWrite) &&
5577 			    rc == X86EMUL_PROPAGATE_FAULT &&
5578 			    ctxt->exception.vector == PF_VECTOR)
5579 				ctxt->exception.error_code |= PFERR_WRITE_MASK;
5580 			goto done;
5581 		}
5582 	}
5583 	/* Copy full 64-bit value for CMPXCHG8B.  */
5584 	ctxt->dst.orig_val64 = ctxt->dst.val64;
5585 
5586 special_insn:
5587 
5588 	if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5589 		rc = emulator_check_intercept(ctxt, ctxt->intercept,
5590 					      X86_ICPT_POST_MEMACCESS);
5591 		if (rc != X86EMUL_CONTINUE)
5592 			goto done;
5593 	}
5594 
5595 	if (ctxt->rep_prefix && (ctxt->d & String))
5596 		ctxt->eflags |= X86_EFLAGS_RF;
5597 	else
5598 		ctxt->eflags &= ~X86_EFLAGS_RF;
5599 
5600 	if (ctxt->execute) {
5601 		if (ctxt->d & Fastop)
5602 			rc = fastop(ctxt, ctxt->fop);
5603 		else
5604 			rc = ctxt->execute(ctxt);
5605 		if (rc != X86EMUL_CONTINUE)
5606 			goto done;
5607 		goto writeback;
5608 	}
5609 
5610 	if (ctxt->opcode_len == 2)
5611 		goto twobyte_insn;
5612 	else if (ctxt->opcode_len == 3)
5613 		goto threebyte_insn;
5614 
5615 	switch (ctxt->b) {
5616 	case 0x70 ... 0x7f: /* jcc (short) */
5617 		if (test_cc(ctxt->b, ctxt->eflags))
5618 			rc = jmp_rel(ctxt, ctxt->src.val);
5619 		break;
5620 	case 0x8d: /* lea r16/r32, m */
5621 		ctxt->dst.val = ctxt->src.addr.mem.ea;
5622 		break;
5623 	case 0x90 ... 0x97: /* nop / xchg reg, rax */
5624 		if (ctxt->dst.addr.reg == reg_rmw(ctxt, VCPU_REGS_RAX))
5625 			ctxt->dst.type = OP_NONE;
5626 		else
5627 			rc = em_xchg(ctxt);
5628 		break;
5629 	case 0x98: /* cbw/cwde/cdqe */
5630 		switch (ctxt->op_bytes) {
5631 		case 2: ctxt->dst.val = (s8)ctxt->dst.val; break;
5632 		case 4: ctxt->dst.val = (s16)ctxt->dst.val; break;
5633 		case 8: ctxt->dst.val = (s32)ctxt->dst.val; break;
5634 		}
5635 		break;
5636 	case 0xcc:		/* int3 */
5637 		rc = emulate_int(ctxt, 3);
5638 		break;
5639 	case 0xcd:		/* int n */
5640 		rc = emulate_int(ctxt, ctxt->src.val);
5641 		break;
5642 	case 0xce:		/* into */
5643 		if (ctxt->eflags & X86_EFLAGS_OF)
5644 			rc = emulate_int(ctxt, 4);
5645 		break;
5646 	case 0xe9: /* jmp rel */
5647 	case 0xeb: /* jmp rel short */
5648 		rc = jmp_rel(ctxt, ctxt->src.val);
5649 		ctxt->dst.type = OP_NONE; /* Disable writeback. */
5650 		break;
5651 	case 0xf4:              /* hlt */
5652 		ctxt->ops->halt(ctxt);
5653 		break;
5654 	case 0xf5:	/* cmc */
5655 		/* complement carry flag from eflags reg */
5656 		ctxt->eflags ^= X86_EFLAGS_CF;
5657 		break;
5658 	case 0xf8: /* clc */
5659 		ctxt->eflags &= ~X86_EFLAGS_CF;
5660 		break;
5661 	case 0xf9: /* stc */
5662 		ctxt->eflags |= X86_EFLAGS_CF;
5663 		break;
5664 	case 0xfc: /* cld */
5665 		ctxt->eflags &= ~X86_EFLAGS_DF;
5666 		break;
5667 	case 0xfd: /* std */
5668 		ctxt->eflags |= X86_EFLAGS_DF;
5669 		break;
5670 	default:
5671 		goto cannot_emulate;
5672 	}
5673 
5674 	if (rc != X86EMUL_CONTINUE)
5675 		goto done;
5676 
5677 writeback:
5678 	if (ctxt->d & SrcWrite) {
5679 		BUG_ON(ctxt->src.type == OP_MEM || ctxt->src.type == OP_MEM_STR);
5680 		rc = writeback(ctxt, &ctxt->src);
5681 		if (rc != X86EMUL_CONTINUE)
5682 			goto done;
5683 	}
5684 	if (!(ctxt->d & NoWrite)) {
5685 		rc = writeback(ctxt, &ctxt->dst);
5686 		if (rc != X86EMUL_CONTINUE)
5687 			goto done;
5688 	}
5689 
5690 	/*
5691 	 * restore dst type in case the decoding will be reused
5692 	 * (happens for string instruction )
5693 	 */
5694 	ctxt->dst.type = saved_dst_type;
5695 
5696 	if ((ctxt->d & SrcMask) == SrcSI)
5697 		string_addr_inc(ctxt, VCPU_REGS_RSI, &ctxt->src);
5698 
5699 	if ((ctxt->d & DstMask) == DstDI)
5700 		string_addr_inc(ctxt, VCPU_REGS_RDI, &ctxt->dst);
5701 
5702 	if (ctxt->rep_prefix && (ctxt->d & String)) {
5703 		unsigned int count;
5704 		struct read_cache *r = &ctxt->io_read;
5705 		if ((ctxt->d & SrcMask) == SrcSI)
5706 			count = ctxt->src.count;
5707 		else
5708 			count = ctxt->dst.count;
5709 		register_address_increment(ctxt, VCPU_REGS_RCX, -count);
5710 
5711 		if (!string_insn_completed(ctxt)) {
5712 			/*
5713 			 * Re-enter guest when pio read ahead buffer is empty
5714 			 * or, if it is not used, after each 1024 iteration.
5715 			 */
5716 			if ((r->end != 0 || reg_read(ctxt, VCPU_REGS_RCX) & 0x3ff) &&
5717 			    (r->end == 0 || r->end != r->pos)) {
5718 				/*
5719 				 * Reset read cache. Usually happens before
5720 				 * decode, but since instruction is restarted
5721 				 * we have to do it here.
5722 				 */
5723 				ctxt->mem_read.end = 0;
5724 				writeback_registers(ctxt);
5725 				return EMULATION_RESTART;
5726 			}
5727 			goto done; /* skip rip writeback */
5728 		}
5729 		ctxt->eflags &= ~X86_EFLAGS_RF;
5730 	}
5731 
5732 	ctxt->eip = ctxt->_eip;
5733 	if (ctxt->mode != X86EMUL_MODE_PROT64)
5734 		ctxt->eip = (u32)ctxt->_eip;
5735 
5736 done:
5737 	if (rc == X86EMUL_PROPAGATE_FAULT) {
5738 		if (KVM_EMULATOR_BUG_ON(ctxt->exception.vector > 0x1f, ctxt))
5739 			return EMULATION_FAILED;
5740 		ctxt->have_exception = true;
5741 	}
5742 	if (rc == X86EMUL_INTERCEPTED)
5743 		return EMULATION_INTERCEPTED;
5744 
5745 	if (rc == X86EMUL_CONTINUE)
5746 		writeback_registers(ctxt);
5747 
5748 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
5749 
5750 twobyte_insn:
5751 	switch (ctxt->b) {
5752 	case 0x09:		/* wbinvd */
5753 		(ctxt->ops->wbinvd)(ctxt);
5754 		break;
5755 	case 0x08:		/* invd */
5756 	case 0x0d:		/* GrpP (prefetch) */
5757 	case 0x18:		/* Grp16 (prefetch/nop) */
5758 	case 0x1f:		/* nop */
5759 		break;
5760 	case 0x20: /* mov cr, reg */
5761 		ctxt->dst.val = ops->get_cr(ctxt, ctxt->modrm_reg);
5762 		break;
5763 	case 0x21: /* mov from dr to reg */
5764 		ops->get_dr(ctxt, ctxt->modrm_reg, &ctxt->dst.val);
5765 		break;
5766 	case 0x40 ... 0x4f:	/* cmov */
5767 		if (test_cc(ctxt->b, ctxt->eflags))
5768 			ctxt->dst.val = ctxt->src.val;
5769 		else if (ctxt->op_bytes != 4)
5770 			ctxt->dst.type = OP_NONE; /* no writeback */
5771 		break;
5772 	case 0x80 ... 0x8f: /* jnz rel, etc*/
5773 		if (test_cc(ctxt->b, ctxt->eflags))
5774 			rc = jmp_rel(ctxt, ctxt->src.val);
5775 		break;
5776 	case 0x90 ... 0x9f:     /* setcc r/m8 */
5777 		ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags);
5778 		break;
5779 	case 0xb6 ... 0xb7:	/* movzx */
5780 		ctxt->dst.bytes = ctxt->op_bytes;
5781 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (u8) ctxt->src.val
5782 						       : (u16) ctxt->src.val;
5783 		break;
5784 	case 0xbe ... 0xbf:	/* movsx */
5785 		ctxt->dst.bytes = ctxt->op_bytes;
5786 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (s8) ctxt->src.val :
5787 							(s16) ctxt->src.val;
5788 		break;
5789 	default:
5790 		goto cannot_emulate;
5791 	}
5792 
5793 threebyte_insn:
5794 
5795 	if (rc != X86EMUL_CONTINUE)
5796 		goto done;
5797 
5798 	goto writeback;
5799 
5800 cannot_emulate:
5801 	return EMULATION_FAILED;
5802 }
5803 
5804 void emulator_invalidate_register_cache(struct x86_emulate_ctxt *ctxt)
5805 {
5806 	invalidate_registers(ctxt);
5807 }
5808 
5809 void emulator_writeback_register_cache(struct x86_emulate_ctxt *ctxt)
5810 {
5811 	writeback_registers(ctxt);
5812 }
5813 
5814 bool emulator_can_use_gpa(struct x86_emulate_ctxt *ctxt)
5815 {
5816 	if (ctxt->rep_prefix && (ctxt->d & String))
5817 		return false;
5818 
5819 	if (ctxt->d & TwoMemOp)
5820 		return false;
5821 
5822 	return true;
5823 }
5824